feat(mcp): raise instead of returning error objects in the Prowler App tools (#12532)

mcp_server/changelog.d/mcp-app-tool-failures.added.md

@@ -0,0 +1 @@
+Prowler App tools now report a failure as an MCP tool execution error (`isError: true`, explanation in `content`) instead of as a successful result carrying an `{"error": ...}` object, which clients and models read as a success

mcp_server/changelog.d/mcp-compliance-scan-or-provider.changed.md

@@ -0,0 +1 @@
+`prowler_get_compliance_framework_state_details` now rejects a call that passes both `scan_id` and `provider_id` instead of silently ignoring the provider, which could report on a scan belonging to a different provider than the one that was asked about

mcp_server/prowler_mcp_server/lib/errors.py

@@ -19,10 +19,17 @@ class ProwlerAPIError(Exception):
     Attributes:
         status_code: HTTP status the API answered with
         detail: JSON:API `errors[0].detail`, None when there is none to trust
+        payload: Parsed JSON body, for a tool that has to read the answer rather
+            than only report it
     """
 
     def __init__(
-        self, message: str, status_code: int, *, detail: str | None = None
+        self,
+        message: str,
+        status_code: int,
+        *,
+        detail: str | None = None,
+        payload: dict[str, Any] | None = None,
     ) -> None:
         super().__init__(message)
         self.status_code: int = status_code
@@ -32,6 +39,12 @@ class ProwlerAPIError(Exception):
         # that must never be repeated to a model -- and None for a 5xx, see
         # `jsonapi_detail`.
         self.detail: str | None = detail
+        # Not every error status means the request failed: Prowler answers 404
+        # with the result itself when a query ran and matched nothing. A tool
+        # reads this to tell such an answer apart from a real failure. It is the
+        # upstream body, so it is read structurally and never relayed as text --
+        # `detail` above is the only part of it that may be repeated to a model.
+        self.payload: dict[str, Any] | None = payload
 
 
 class ProwlerAPIUnreachable(Exception):
@@ -71,6 +84,10 @@ class InvalidArgument(ValueError):
     """An argument this server rejected before any request went out."""
 
 
+class CredentialError(Exception):
+    """The credential the caller sent is missing, malformed or expired."""
+
+
 # ------------------------------------------------------------------- messages
 
 
@@ -154,6 +171,16 @@ def _describe_failure(exc: BaseException) -> str | None:
             "current state before sending it again."
         )
 
+    if isinstance(exc, CredentialError):
+        # Not an argument problem, so it is worth saying that plainly: the
+        # answer is a credential the user has to fix, not another attempt.
+        return (
+            f"This request carried no usable credential: {exc}. Retrying or "
+            "changing the arguments will not help -- the client has to send an "
+            "'Authorization: Bearer <token>' header holding a valid Prowler API "
+            "key or an unexpired JWT."
+        )
+
     if isinstance(exc, ProwlerAPIUnreachable):
         # The only failure a model can turn into a duplicate write by repeating.
         return (

mcp_server/prowler_mcp_server/lib/types.py

@@ -0,0 +1,18 @@
+"""Argument types shared by every tool in this server."""
+
+from typing import Annotated
+
+from pydantic import StringConstraints
+
+# The identifiers tools take -- a scan UUID, a query id, a Jira project key --
+# are required because there is nothing sensible to do without them. A model
+# that does not have one to hand tends to send an empty string rather than omit
+# the argument, and an empty string is not caught by "required": it travels into
+# a URL path or a request body and comes back as a 404 or an opaque API error
+# ("This field may not be blank") that says nothing about which argument was at
+# fault. Rejecting it here names the argument instead, and `minLength` puts the
+# constraint in the tool schema so a client can see it before calling.
+#
+# Whitespace is stripped first, so " abc " is accepted as "abc" and "   " is
+# rejected like "".
+NonBlankStr = Annotated[str, StringConstraints(strip_whitespace=True, min_length=1)]

mcp_server/prowler_mcp_server/prowler_app/models/attack_paths.py

@@ -354,6 +354,14 @@ class AttackPathQueryResult(MinimalSerializerMixin, BaseModel):
     relationships: list[AttackPathsGraphRelationship] = Field(
         default_factory=list, description="Relationships connecting the nodes"
     )
+    # A graph with nothing in it serializes to `{}`, since the mixin drops empty
+    # lists. That reads as an answer that went missing rather than as the finding
+    # it is -- the query ran and this account has no such attack path -- so the
+    # empty case carries a sentence saying so.
+    message: str | None = Field(
+        default=None,
+        description="Present only when the query matched nothing, to say the query ran and found no attack path rather than leaving an empty result to interpret",
+    )
 
     @classmethod
     def from_api_response(
@@ -368,7 +376,15 @@ class AttackPathQueryResult(MinimalSerializerMixin, BaseModel):
         Returns:
             AttackPathQueryResult with parsed data and summary
         """
-        attributes = response.get("data", {}).get("attributes")
+        data = response.get("data")
+        attributes = data.get("attributes") if data is not None else None
+        # Prowler spells a graph with nothing in it either as empty lists or as
+        # a null `attributes`. Both say the same thing -- the query ran and
+        # matched nothing -- so the null reads as the empty graph it stands for
+        # instead of crashing the parse.
+        if attributes is None:
+            attributes = {}
+
         nodes_data = attributes.get("nodes", [])
         relationships_data = attributes.get("relationships", [])
 

mcp_server/prowler_mcp_server/prowler_app/models/providers.py

@@ -2,7 +2,7 @@
 
 from typing import Any, Literal
 
-from pydantic import BaseModel
+from pydantic import BaseModel, ConfigDict, Field
 
 from prowler_mcp_server.prowler_app.models.base import MinimalSerializerMixin
 
@@ -104,6 +104,29 @@ class ProvidersListResponse(BaseModel):
         )
 
 
+class ProviderDeletionResult(MinimalSerializerMixin, BaseModel):
+    """Outcome of a provider deletion.
+
+    Prowler deletes a provider in a background task, so the answer is not always
+    a finished deletion. A deletion that never started is raised as an error
+    instead of being reported here: this model only describes a deletion Prowler
+    accepted and began.
+    """
+
+    model_config = ConfigDict(frozen=True)
+
+    status: Literal["deleted", "in_progress"] = Field(
+        description="Outcome of the deletion: 'deleted' when Prowler finished removing the provider, 'in_progress' when the background task was accepted and is still running, which is normal for a provider with many scans and findings"
+    )
+    task_id: str | None = Field(
+        default=None,
+        description="UUIDv4 of the background deletion task, present when the deletion did not finish within the polling window so its state can be checked later",
+    )
+    message: str = Field(
+        description="Human-readable description of what happened and what to do next"
+    )
+
+
 class ProviderConnectionStatus(MinimalSerializerMixin, BaseModel):
     """Result of provider connection operation."""
 

mcp_server/prowler_mcp_server/prowler_app/models/scans.py

@@ -191,18 +191,18 @@ class ScansListResponse(BaseModel):
 
 
 class ScanCreationResult(MinimalSerializerMixin, BaseModel):
-    """Result of scan creation operation.
+    """Result of a scan creation that succeeded.
 
-    Used by trigger_scan() to communicate the outcome of scan creation.
+    Used by trigger_scan(). A scan that was not created leaves the tool as an
-    Status indicates whether scan was created successfully or failed.
+    error instead of being reported here, so this model only ever describes a
+    scan that exists -- which is why it carries no success flag: a field with
+    one reachable value says nothing, and inviting a reader to branch on it
+    suggests there is a failure shape to look for here. There is not; the
+    failure is the error.
     """
 
-    scan: DetailedScan | None = Field(
+    scan: DetailedScan = Field(
-        default=None,
+        description="Detailed information about the scan that was created"
-        description="Detailed scan information if creation succeeded, None otherwise",
-    )
-    status: Literal["success", "failed"] = Field(
-        description="Outcome of scan creation: success (scan created successfully) or failed (error)"
     )
     message: str = Field(
         description="Human-readable message describing the scan creation result"
@@ -210,13 +210,26 @@ class ScanCreationResult(MinimalSerializerMixin, BaseModel):
 
 
 class ScheduleCreationResult(MinimalSerializerMixin, BaseModel):
-    """Result of async schedule creation operation.
+    """Result of a daily schedule creation that succeeded.
 
-    Used by schedule_daily_scan() to communicate scheduling outcome.
+    Used by schedule_daily_scan(). Prowler commits the schedule inside the
+    request that creates it, so an answer means it exists; a provider that
+    already has one is refused with a 409 and leaves the tool as an error. That
+    leaves nothing for a success flag to distinguish, so there is none.
     """
 
-    scheduled: bool = Field(
+    first_run_state: (
-        description="Whether the daily scan schedule was created successfully"
+        Literal[
+            "available", "scheduled", "executing", "completed", "failed", "cancelled"
+        ]
+        | None
+    ) = Field(
+        default=None,
+        description=(
+            "State of the first scan Prowler starts immediately alongside the schedule. "
+            "This describes that one run, not the recurring schedule, which stands "
+            "regardless of it"
+        ),
     )
     message: str = Field(
         description="Human-readable message describing the scheduling result"

mcp_server/prowler_mcp_server/prowler_app/server.py

@@ -3,7 +3,7 @@ from fastmcp import FastMCP
 from prowler_mcp_server.prowler_app.utils.tool_loader import load_all_tools
 
 # Initialize MCP server
-app_mcp_server = FastMCP("prowler-app")
+app_mcp_server = FastMCP("prowler-app", mask_error_details=True)
 
 # Auto-discover and load all tools from the tools package
 load_all_tools(app_mcp_server)

mcp_server/prowler_mcp_server/prowler_app/tools/attack_paths.py

@@ -7,8 +7,11 @@ through cloud infrastructure relationships.
 
 from typing import Any, Literal
 
+from fastmcp.exceptions import ToolError
 from pydantic import Field
 
+from prowler_mcp_server.lib.errors import ProwlerAPIError
+from prowler_mcp_server.lib.types import NonBlankStr
 from prowler_mcp_server.prowler_app.models.attack_paths import (
     AttackPathCartographySchema,
     AttackPathQuery,
@@ -76,7 +79,6 @@ class AttackPathsTools(BaseTool):
         2. Use prowler_list_attack_paths_queries to see available queries for a scan
         3. Use prowler_run_attack_paths_query to execute analysis
         """
-        try:
         # Validate pagination
         self.api_client.validate_page_size(page_size)
 
@@ -106,20 +108,18 @@ class AttackPathsTools(BaseTool):
         )
 
         return simplified_response.model_dump()
-        except Exception as e:
-            self.logger.error(f"Failed to list attack paths scans: {e}")
-            return {"error": f"Failed to list attack paths scans: {str(e)}"}
 
     async def list_attack_paths_queries(
         self,
-        scan_id: str = Field(
+        scan_id: NonBlankStr = Field(
-            description="UUID of a COMPLETED attack paths scan. Use `prowler_list_attack_paths_scans` with state=['completed'] to find scan IDs"
+            description="UUID of a COMPLETED attack paths scan, as returned by `prowler_list_attack_paths_scans` with state=['completed']. This is NOT a regular scan ID: an ID from `prowler_search_scans` or `prowler_get_scan` names a different resource and is rejected here"
         ),
     ) -> list[dict[str, Any]]:
         """Discover available Attack Paths queries for a completed scan.
 
         IMPORTANT: The scan must be in 'completed' state to list queries.
-        Queries are provider-specific
+        Attack Paths covers AWS providers only, so only an AWS provider has an
+        Attack Paths scan to name here, and every query is an AWS one.
 
         Each query includes:
         - id: Query identifier to use with run_attack_paths_query
@@ -141,23 +141,32 @@ class AttackPathsTools(BaseTool):
             api_response = await self.api_client.get(
                 f"/attack-paths-scans/{scan_id}/queries"
             )
+        except ProwlerAPIError as e:
+            # A 404 here is Prowler failing to resolve `scan_id` to an Attack
+            # Paths scan, and its own reason for it -- a bare "Not found." --
+            # does not say what kind of ID it was looking for. The mistake it
+            # stands for is a regular scan ID: an Attack Paths scan is a separate
+            # resource with IDs of its own, and Prowler only creates one for an
+            # AWS provider, so a scan of any other provider has none to pass.
+            #
+            # The endpoint answers 404 for a second thing -- a provider type with
+            # no query catalog -- but that one cannot happen: a scan only exists
+            # where Attack Paths runs, which is AWS, and AWS has a catalog.
+            if e.status_code == 404:
+                raise self._unknown_scan_error(scan_id)
+            raise
 
         return [
             AttackPathQuery.from_api_response(query).model_dump()
             for query in api_response.get("data", [])
         ]
-        except Exception as e:
-            self.logger.error(
-                f"Failed to list attack paths queries for scan {scan_id}: {e}"
-            )
-            return [{"error": f"Failed to list attack paths queries: {str(e)}"}]
 
     async def run_attack_paths_query(
         self,
-        scan_id: str = Field(
+        scan_id: NonBlankStr = Field(
             description="UUID of a COMPLETED attack paths scan. The scan must be in 'completed' state"
         ),
-        query_id: str = Field(
+        query_id: NonBlankStr = Field(
             description="Query ID to execute (e.g., 'aws-internet-exposed-ec2-sensitive-s3-access'). Use `prowler_list_attack_paths_queries` to discover available queries"
         ),
         parameters: dict[str, str] = Field(
@@ -198,7 +207,6 @@ class AttackPathsTools(BaseTool):
         3. Execute this tool with appropriate parameters
         4. Analyze the returned graph for security insights
         """
-        try:
         # Build the request payload following JSON:API format
         request_data: dict[str, Any] = {
             "data": {
@@ -213,24 +221,47 @@ class AttackPathsTools(BaseTool):
         if parameters:
             request_data["data"]["attributes"]["parameters"] = parameters
 
+        try:
             api_response = await self.api_client.post(
                 f"/attack-paths-scans/{scan_id}/queries/run",
                 json_data=request_data,
             )
+        except ProwlerAPIError as e:
+            # Prowler answers a query that matched nothing with 404 and the empty
+            # result as the body. That is an answer -- this account has no such
+            # attack path, which is the good outcome -- so it is returned rather
+            # than raised: reporting it as a failure invites a retry of a call
+            # whose arguments were right, and hides a clean result.
+            if e.status_code == 404 and isinstance(e.payload, dict):
+                if "data" in e.payload:
+                    api_response = e.payload
+                else:
+                    # No result body, so `scan_id` did not resolve to an Attack
+                    # Paths scan. An unknown query_id is a 400, not this.
+                    raise self._unknown_scan_error(scan_id)
+            else:
+                raise
 
         # Parse the response
         query_result = AttackPathQueryResult.from_api_response(api_response)
 
-            return query_result.model_dump()
+        if not query_result.nodes:
-        except Exception as e:
+            query_result = query_result.model_copy(
-            self.logger.error(
+                update={
-                f"Failed to run attack paths query '{query_id}' on scan {scan_id}: {e}"
+                    "message": (
+                        f"The query '{query_id}' ran against scan {scan_id} and matched "
+                        "nothing, so this provider has no attack path of that shape. "
+                        "The scan and the query ID were both valid; running it again "
+                        "will return the same thing."
                     )
-            return {"error": f"Failed to run attack paths query '{query_id}': {str(e)}"}
+                }
+            )
+
+        return query_result.model_dump()
 
     async def get_attack_paths_cartography_schema(
         self,
-        scan_id: str = Field(
+        scan_id: NonBlankStr = Field(
             description="UUID of a COMPLETED attack paths scan. Use `prowler_list_attack_paths_scans` with state=['completed'] to find scan IDs"
         ),
     ) -> dict[str, Any]:
@@ -262,18 +293,43 @@ class AttackPathsTools(BaseTool):
             api_response = await self.api_client.get(
                 f"/attack-paths-scans/{scan_id}/schema"
             )
+        except ProwlerAPIError as e:
+            # Two 404s again, told apart by whether Prowler wrote a JSON:API
+            # error. Absent means the scan resolved and its graph simply records
+            # no Cartography module, so the ID is not the thing to change.
+            if e.status_code == 404:
+                if e.detail is None:
+                    raise ToolError(
+                        f"Scan {scan_id} has no Cartography schema recorded, so there is "
+                        "nothing to write custom queries against. Use "
+                        "prowler_list_attack_paths_queries for the ready-made queries of "
+                        "this scan, which do not need the schema."
+                    )
+                else:
+                    raise self._unknown_scan_error(scan_id)
+            raise
 
         schema = AttackPathCartographySchema.from_api_response(api_response)
 
-            schema_content = await self.api_client.fetch_external_url(
+        schema_content = await self.api_client.fetch_external_url(schema.raw_schema_url)
-                schema.raw_schema_url
-            )
 
-            return schema.model_copy(
+        return schema.model_copy(update={"schema_content": schema_content}).model_dump()
-                update={"schema_content": schema_content}
+
-            ).model_dump()
+    # Private helper methods
-        except Exception as e:
+
-            self.logger.error(
+    @staticmethod
-                f"Failed to get cartography schema for scan {scan_id}: {e}"
+    def _unknown_scan_error(scan_id: str) -> ToolError:
+        """Describe a scan ID Prowler could not resolve to an Attack Paths scan.
+
+        Returns:
+            The ``ToolError`` for the caller to raise. Built without a ``from``
+            clause on purpose: the sentence is the final word, not a wrapper
+            around the API's.
+        """
+        return ToolError(
+            f"Prowler has no Attack Paths scan with ID {scan_id}. These are a "
+            "different resource from regular scans and only exist for AWS "
+            "providers, so an ID from prowler_search_scans or prowler_get_scan "
+            "never resolves here. Use prowler_list_attack_paths_scans to get an "
+            "ID these tools take."
         )
-            return {"error": f"Failed to get cartography schema: {str(e)}"}

mcp_server/prowler_mcp_server/prowler_app/tools/compliance.py

@@ -6,8 +6,11 @@ across all cloud providers.
 
 from typing import Any
 
+from fastmcp.exceptions import ToolError
 from pydantic import Field
 
+from prowler_mcp_server.lib.errors import InvalidArgument
+from prowler_mcp_server.lib.types import NonBlankStr
 from prowler_mcp_server.prowler_app.models.compliance import (
     ComplianceFrameworksListResponse,
     ComplianceRequirementAttributesListResponse,
@@ -34,7 +37,7 @@ class ComplianceTools(BaseTool):
             The scan_id of the latest completed scan for the provider.
 
         Raises:
-            ValueError: If no completed scans are found for the provider.
+            ToolError: If no completed scans are found for the provider
         """
         scan_params = {
             "filter[provider]": provider_id,
@@ -48,7 +51,7 @@ class ComplianceTools(BaseTool):
 
         scans_data = scans_response.get("data", [])
         if not scans_data:
-            raise ValueError(
+            raise ToolError(
                 f"No completed scans found for provider {provider_id}. "
                 "Run a scan first using prowler_trigger_scan."
             )
@@ -93,18 +96,15 @@ class ComplianceTools(BaseTool):
         2. Use prowler_get_compliance_framework_state_details with a specific compliance_id to see which requirements failed
         """
         if not scan_id and not provider_id:
-            return {
+            raise InvalidArgument(
-                "error": "Either scan_id or provider_id must be provided. Use prowler_search_providers to find provider IDs or prowler_list_scans to find scan IDs."
+                "Either scan_id or provider_id must be provided. Use prowler_search_providers to find provider IDs or prowler_list_scans to find scan IDs."
-            }
+            )
         elif scan_id and provider_id:
-            return {
+            raise InvalidArgument(
-                "error": "Provide either scan_id or provider_id, not both. To get compliance data for a specific scan, use scan_id. To get data for the latest scan of a provider, use provider_id."
+                "Provide either scan_id or provider_id, not both. To get compliance data for a specific scan, use scan_id. To get data for the latest scan of a provider, use provider_id."
-            }
+            )
         elif not scan_id and provider_id:
-            try:
             scan_id = await self._get_latest_scan_id_for_provider(provider_id)
-            except ValueError as e:
-                return {"error": str(e)}
 
         params: dict[str, Any] = {"filter[scan_id]": scan_id}
 
@@ -253,16 +253,16 @@ class ComplianceTools(BaseTool):
 
     async def get_compliance_framework_state_details(
         self,
-        compliance_id: str = Field(
+        compliance_id: NonBlankStr = Field(
             description="Compliance framework ID to get details for (e.g., 'cis_1.5_aws', 'pci_dss_v4.0_aws'). You can get compliance IDs from prowler_get_compliance_overview or consulting Prowler Hub/Prowler Documentation that you can also find in form of tools in this MCP Server",
         ),
         scan_id: str | None = Field(
             default=None,
-            description="UUID of a specific scan to get compliance data for. Required if provider_id is not specified.",
+            description="UUID of a specific scan to get compliance data for. Required if provider_id is not specified. Do not pass it together with provider_id.",
         ),
         provider_id: str | None = Field(
             default=None,
-            description="Prowler's internal UUID (v4) for a specific provider. If provided without scan_id, the tool will automatically find the latest completed scan for this provider. Use `prowler_search_providers` tool to find provider IDs.",
+            description="Prowler's internal UUID (v4) for a specific provider. The tool will automatically find the latest completed scan for this provider. Use `prowler_search_providers` tool to find provider IDs. Do not pass it together with scan_id.",
         ),
     ) -> dict[str, Any]:
         """Get detailed requirement-level breakdown for a specific compliance framework.
@@ -283,8 +283,8 @@ class ComplianceTools(BaseTool):
            - Use prowler_get_finding_details with these finding IDs for more details and remediation guidance
 
         Default behavior:
-        - Requires either scan_id OR provider_id
+        - Requires exactly one of scan_id OR provider_id; providing both is rejected
-        - With provider_id (no scan_id): Automatically finds the latest completed scan for that provider
+        - With provider_id: Automatically finds the latest completed scan for that provider
         - With scan_id: Uses that specific scan's compliance data
         - Only shows failed requirements with their associated failed finding IDs
 
@@ -293,21 +293,22 @@ class ComplianceTools(BaseTool):
         2. Use this tool with the compliance_id to see failed requirements and their finding IDs
         3. Use prowler_get_finding_details with the finding IDs to get remediation guidance
         """
-        # Validate that either scan_id or provider_id is provided
+        # Exactly one of the two: taking scan_id and ignoring provider_id would
+        # answer for whatever provider that scan belongs to, which is not
+        # necessarily the one the caller named.
         if not scan_id and not provider_id:
-            return {
+            raise InvalidArgument(
-                "error": "Either scan_id or provider_id must be provided. Use prowler_search_providers to find provider IDs or prowler_list_scans to find scan IDs."
+                "Either scan_id or provider_id must be provided. Use prowler_search_providers to find provider IDs or prowler_list_scans to find scan IDs."
-            }
+            )
+        elif scan_id and provider_id:
+            raise InvalidArgument(
+                "Provide either scan_id or provider_id, not both. To get compliance data for a specific scan, use scan_id. To get data for the latest scan of a provider, use provider_id."
+            )
 
         # Resolve provider_id to latest scan_id if needed
         resolved_scan_id = scan_id
         if not scan_id and provider_id:
-            try:
+            resolved_scan_id = await self._get_latest_scan_id_for_provider(provider_id)
-                resolved_scan_id = await self._get_latest_scan_id_for_provider(
-                    provider_id
-                )
-            except ValueError as e:
-                return {"error": str(e)}
 
         # Build params for requirements endpoint
         params: dict[str, Any] = {

mcp_server/prowler_mcp_server/prowler_app/tools/finding_groups.py

@@ -6,8 +6,10 @@ This module provides read-only tools for finding group triage and drill-downs.
 from typing import Any, Literal
 from urllib.parse import quote
 
+from fastmcp.exceptions import ToolError
 from pydantic import Field
 
+from prowler_mcp_server.lib.types import NonBlankStr
 from prowler_mcp_server.prowler_app.models.finding_groups import (
     DetailedFindingGroup,
     FindingGroupResourcesListResponse,
@@ -236,7 +238,6 @@ class FindingGroupsTools(BaseTool):
         prowler_get_finding_group_details for complete counters or
         prowler_list_finding_group_resources to drill into affected resources.
         """
-        try:
         self.api_client.validate_page_size(page_size)
         date_range, params = self._base_date_params(date_from, date_to)
         endpoint = self._group_endpoint(date_range)
@@ -273,13 +274,10 @@ class FindingGroupsTools(BaseTool):
         api_response = await self.api_client.get(endpoint, params=clean_params)
         response = FindingGroupsListResponse.from_api_response(api_response)
         return response.model_dump()
-        except Exception as e:
-            self.logger.error(f"Error listing finding groups: {e}")
-            return {"error": str(e), "status": "failed"}
 
     async def get_finding_group_details(
         self,
-        check_id: str = Field(
+        check_id: NonBlankStr = Field(
             description="Public check ID that identifies the finding group. This is not a UUID."
         ),
         date_from: str | None = Field(
@@ -297,7 +295,6 @@ class FindingGroupsTools(BaseTool):
         or historical data when dates are provided. Fully muted groups are
         included by default so accepted risk does not look like a missing group.
         """
-        try:
         date_range, params = self._base_date_params(date_from, date_to)
         endpoint = self._group_endpoint(date_range)
 
@@ -316,20 +313,19 @@ class FindingGroupsTools(BaseTool):
         data = api_response.get("data", [])
 
         if not data:
-                return {
+            # No `from`: this names the check and the tool that lists valid ones,
-                    "error": f"Finding group '{check_id}' not found.",
+            # neither of which the shared classifier can know.
-                    "status": "not_found",
+            raise ToolError(
-                }
+                f"No finding group exists for check '{check_id}' in this scan. Use "
+                "prowler_list_finding_groups to see the checks that have findings."
+            )
 
         group = DetailedFindingGroup.from_api_response(data[0])
         return group.model_dump()
-        except Exception as e:
-            self.logger.error(f"Error getting finding group details: {e}")
-            return {"error": str(e), "status": "failed"}
 
     async def list_finding_group_resources(
         self,
-        check_id: str = Field(
+        check_id: NonBlankStr = Field(
             description="Public check ID that identifies the finding group. This is not a UUID."
         ),
         provider: list[str] = Field(
@@ -426,7 +422,6 @@ class FindingGroupsTools(BaseTool):
         `finding_id`. Use `prowler_get_finding_details(finding_id)` to
         retrieve complete remediation guidance for a specific resource finding.
         """
-        try:
         self.api_client.validate_page_size(page_size)
         date_range, params = self._base_date_params(date_from, date_to)
         endpoint = self._resource_endpoint(check_id, date_range)
@@ -465,6 +460,3 @@ class FindingGroupsTools(BaseTool):
         api_response = await self.api_client.get(endpoint, params=clean_params)
         response = FindingGroupResourcesListResponse.from_api_response(api_response)
         return response.model_dump()
-        except Exception as e:
-            self.logger.error(f"Error listing finding group resources: {e}")
-            return {"error": str(e), "status": "failed"}

mcp_server/prowler_mcp_server/prowler_app/tools/findings.py

@@ -8,6 +8,7 @@ from typing import Any, Literal
 
 from pydantic import Field
 
+from prowler_mcp_server.lib.types import NonBlankStr
 from prowler_mcp_server.prowler_app.models.findings import (
     DetailedFinding,
     FindingsListResponse,
@@ -180,7 +181,7 @@ class FindingsTools(BaseTool):
 
     async def get_finding_details(
         self,
-        finding_id: str = Field(
+        finding_id: NonBlankStr = Field(
             description="UUID of the finding to retrieve (must be a valid UUID format, e.g., '019ac0d6-90d5-73e9-9acf-c22e256f1bac'). Returns an error if the finding ID is invalid or not found."
         ),
     ) -> dict[str, Any]:

mcp_server/prowler_mcp_server/prowler_app/tools/integrations.py

@@ -9,8 +9,11 @@ This module provides tools for managing where Prowler sends its results, includi
 import json
 from typing import Any
 
+from fastmcp.exceptions import ToolError
 from pydantic import Field
 
+from prowler_mcp_server.lib.errors import CredentialError, InvalidArgument
+from prowler_mcp_server.lib.types import NonBlankStr
 from prowler_mcp_server.prowler_app.models.integrations import (
     DetailedIntegration,
     IntegrationConnectionStatus,
@@ -126,7 +129,7 @@ class IntegrationsTools(BaseTool):
 
     async def get_integration(
         self,
-        integration_id: str = Field(
+        integration_id: NonBlankStr = Field(
             description="UUID of the integration to retrieve. Must be a valid UUID format (e.g., '019ac0d6-90d5-73e9-9acf-c22e256f1bac'). Use prowler_list_integrations to find it."
         ),
     ) -> dict[str, Any]:
@@ -157,7 +160,7 @@ class IntegrationsTools(BaseTool):
 
     async def create_amazon_s3_integration(
         self,
-        bucket_name: str = Field(
+        bucket_name: NonBlankStr = Field(
             description="Name of the S3 bucket where Prowler will upload the scan outputs (CSV, HTML, OCSF JSON and compliance reports)."
         ),
         output_directory: str = Field(
@@ -168,15 +171,15 @@ class IntegrationsTools(BaseTool):
             default=[],
             description="Prowler UUIDs of the providers whose scan outputs are exported to this bucket. Use prowler_search_providers to find them. Leave empty to attach no provider yet.",
         ),
-        role_arn: str | None = Field(
+        role_arn: NonBlankStr | None = Field(
             default=None,
             description="ARN of the IAM role Prowler assumes to write to the bucket (e.g. 'arn:aws:iam::123456789012:role/ProwlerS3Integration'). Recommended over static keys.",
         ),
-        external_id: str | None = Field(
+        external_id: NonBlankStr | None = Field(
             default=None,
             description="External ID required by the trust policy of the assumed role. In Prowler Cloud this is the tenant ID.",
         ),
-        role_session_name: str | None = Field(
+        role_session_name: NonBlankStr | None = Field(
             default=None,
             description="Identifier for the role session, useful to track it in AWS logs. Only letters, digits and the characters =,.@_- are allowed.",
         ),
@@ -184,15 +187,15 @@ class IntegrationsTools(BaseTool):
             default=3600,
             description="Duration of the assumed role session in seconds. Must be between 900 and 43200. Defaults to 3600 when omitted.",
         ),
-        aws_access_key_id: str | None = Field(
+        aws_access_key_id: NonBlankStr | None = Field(
             default=None,
             description="AWS access key ID. Only needed when the Prowler deployment has no ambient AWS credentials.",
         ),
-        aws_secret_access_key: str | None = Field(
+        aws_secret_access_key: NonBlankStr | None = Field(
             default=None,
             description="AWS secret access key. Required when 'aws_access_key_id' is provided.",
         ),
-        aws_session_token: str | None = Field(
+        aws_session_token: NonBlankStr | None = Field(
             default=None,
             description="AWS session token, only for temporary credentials.",
         ),
@@ -244,7 +247,6 @@ class IntegrationsTools(BaseTool):
         """
         self.logger.info(f"Creating Amazon S3 integration for bucket {bucket_name}...")
 
-        try:
         credentials = self._build_aws_credentials(
             role_arn=role_arn,
             external_id=external_id,
@@ -265,13 +267,10 @@ class IntegrationsTools(BaseTool):
             provider_ids=provider_ids,
             enabled=enabled,
         )
-        except Exception as e:
-            self.logger.error(f"Amazon S3 integration creation failed: {e}")
-            return {"error": str(e), "status": "failed"}
 
     async def create_aws_security_hub_integration(
         self,
-        provider_id: str = Field(
+        provider_id: NonBlankStr = Field(
             description="Prowler UUID of the AWS provider whose findings are sent to Security Hub. It must be an AWS provider, and it can only have one Security Hub integration. Use prowler_search_providers with provider_type=['aws'] to find it."
         ),
         send_only_fails: bool = Field(
@@ -282,15 +281,15 @@ class IntegrationsTools(BaseTool):
             default=False,
             description="When true, findings that are no longer present in the latest scan are archived in Security Hub.",
         ),
-        role_arn: str | None = Field(
+        role_arn: NonBlankStr | None = Field(
             default=None,
             description="ARN of a dedicated IAM role Prowler assumes to write to Security Hub. Leave every credential parameter empty to reuse the credentials already stored for the provider, which is the recommended setup.",
         ),
-        external_id: str | None = Field(
+        external_id: NonBlankStr | None = Field(
             default=None,
             description="External ID required by the trust policy of the assumed role.",
         ),
-        role_session_name: str | None = Field(
+        role_session_name: NonBlankStr | None = Field(
             default=None,
             description="Identifier for the role session, useful to track it in AWS logs. Only letters, digits and the characters =,.@_- are allowed.",
         ),
@@ -298,14 +297,14 @@ class IntegrationsTools(BaseTool):
             default=None,
             description="Duration of the assumed role session in seconds. Must be between 900 and 43200. Defaults to 3600 when omitted.",
         ),
-        aws_access_key_id: str | None = Field(
+        aws_access_key_id: NonBlankStr | None = Field(
             default=None, description="AWS access key ID for dedicated credentials."
         ),
-        aws_secret_access_key: str | None = Field(
+        aws_secret_access_key: NonBlankStr | None = Field(
             default=None,
             description="AWS secret access key. Required when 'aws_access_key_id' is provided.",
         ),
-        aws_session_token: str | None = Field(
+        aws_session_token: NonBlankStr | None = Field(
             default=None,
             description="AWS session token, only for temporary credentials.",
         ),
@@ -344,7 +343,6 @@ class IntegrationsTools(BaseTool):
             f"Creating AWS Security Hub integration for provider {provider_id}..."
         )
 
-        try:
         credentials = self._build_aws_credentials(
             role_arn=role_arn,
             external_id=external_id,
@@ -365,19 +363,16 @@ class IntegrationsTools(BaseTool):
             provider_ids=[provider_id],
             enabled=enabled,
         )
-        except Exception as e:
-            self.logger.error(f"AWS Security Hub integration creation failed: {e}")
-            return {"error": str(e), "status": "failed"}
 
     async def create_jira_integration(
         self,
-        domain: str = Field(
+        domain: NonBlankStr = Field(
             description="Atlassian site name, without the '.atlassian.net' suffix. For the site 'https://acme.atlassian.net' the value is 'acme'. Full URLs are accepted and normalized automatically."
         ),
-        user_mail: str = Field(
+        user_mail: NonBlankStr = Field(
             description="Email address of the Atlassian account that owns the API token."
         ),
-        api_token: str = Field(
+        api_token: NonBlankStr = Field(
             description="Atlassian API token, created from the account settings. It needs the 'read:jira-user', 'read:jira-work' and 'write:jira-work' scopes."
         ),
         enabled: bool = Field(
@@ -416,11 +411,8 @@ class IntegrationsTools(BaseTool):
         3. Use prowler_get_jira_issue_types with that project key to pick an issue type
         4. Use prowler_send_findings_to_jira to create the work items
         """
-        try:
         normalized_domain = self._normalize_atlassian_domain(domain)
-            self.logger.info(
+        self.logger.info(f"Creating Jira integration for domain {normalized_domain}...")
-                f"Creating Jira integration for domain {normalized_domain}..."
-            )
 
         return await self._create_integration(
             integration_type="jira",
@@ -434,13 +426,10 @@ class IntegrationsTools(BaseTool):
             provider_ids=[],
             enabled=enabled,
         )
-        except Exception as e:
-            self.logger.error(f"Jira integration creation failed: {e}")
-            return {"error": str(e), "status": "failed"}
 
     async def update_integration(
         self,
-        integration_id: str = Field(
+        integration_id: NonBlankStr = Field(
             description="UUID of the integration to update. Use prowler_list_integrations to find it."
         ),
         enabled: bool | None = Field(
@@ -494,7 +483,6 @@ class IntegrationsTools(BaseTool):
         """
         self.logger.info(f"Updating integration {integration_id}...")
 
-        try:
         current = DetailedIntegration.from_api_response(
             await self._get_integration_raw(integration_id)
         )
@@ -502,11 +490,11 @@ class IntegrationsTools(BaseTool):
 
         if provider_ids is not None:
             if integration_type == "jira":
-                    raise ValueError(
+                raise InvalidArgument(
                     "Jira integrations are tenant-wide and cannot be attached to providers."
                 )
             if integration_type == "aws_security_hub" and len(provider_ids) != 1:
-                    raise ValueError(
+                raise InvalidArgument(
                     "AWS Security Hub integrations must stay attached to exactly one AWS "
                     f"provider, got {len(provider_ids)}. Pass a single provider ID, or use "
                     "prowler_delete_integration to stop sending findings to Security Hub."
@@ -523,7 +511,7 @@ class IntegrationsTools(BaseTool):
 
         if configuration is not None:
             if integration_type == "jira":
-                    raise ValueError(
+                raise InvalidArgument(
                     "Jira integrations do not accept a configuration: it is generated by Prowler. "
                     "Update the credentials instead, or run prowler_test_integration_connection to "
                     "refresh the available projects and issue types."
@@ -547,9 +535,7 @@ class IntegrationsTools(BaseTool):
             }
         }
         if provider_ids is not None:
-                update_body["data"]["relationships"] = _providers_relationship(
+            update_body["data"]["relationships"] = _providers_relationship(provider_ids)
-                    provider_ids
-                )
 
         await self.api_client.patch(
             f"/integrations/{integration_id}", json_data=update_body
@@ -561,14 +547,10 @@ class IntegrationsTools(BaseTool):
             current.provider_ids
         )
         recheck_connection = (
-                credentials is not None
+            credentials is not None or configuration is not None or providers_changed
-                or configuration is not None
-                or providers_changed
         )
         connection_status = (
-                await self._test_connection(integration_id)
+            await self._test_connection(integration_id) if recheck_connection else None
-                if recheck_connection
-                else None
         )
 
         updated = await self._get_integration_raw(integration_id)
@@ -577,13 +559,10 @@ class IntegrationsTools(BaseTool):
                 updated, connection_status
             ).model_dump()
         return DetailedIntegration.from_api_response(updated).model_dump()
-        except Exception as e:
-            self.logger.error(f"Integration update failed: {e}")
-            return {"error": str(e), "status": "failed"}
 
     async def delete_integration(
         self,
-        integration_id: str = Field(
+        integration_id: NonBlankStr = Field(
             description="UUID of the integration to permanently remove. Use prowler_list_integrations to find it."
         ),
     ) -> dict[str, Any]:
@@ -606,22 +585,15 @@ class IntegrationsTools(BaseTool):
         """
         self.logger.info(f"Deleting integration {integration_id}...")
 
-        try:
         await self.api_client.delete(f"/integrations/{integration_id}")
-            return {
+        # No `deleted` flag: an integration that was not deleted leaves this tool
-                "deleted": True,
+        # as an error, so the flag could only ever be True and a reader branching
-                "message": f"Integration {integration_id} deleted successfully",
+        # on it would be looking for a shape that does not exist.
-            }
+        return {"message": f"Integration {integration_id} deleted successfully"}
-        except Exception as e:
-            self.logger.error(f"Integration deletion failed: {e}")
-            return {
-                "deleted": False,
-                "message": f"Integration {integration_id} deletion failed: {str(e)}",
-            }
 
     async def test_integration_connection(
         self,
-        integration_id: str = Field(
+        integration_id: NonBlankStr = Field(
             description="UUID of the integration to check. Use prowler_list_integrations to find it."
         ),
     ) -> dict[str, Any]:
@@ -654,10 +626,10 @@ class IntegrationsTools(BaseTool):
 
     async def get_jira_issue_types(
         self,
-        integration_id: str = Field(
+        integration_id: NonBlankStr = Field(
             description="UUID of the Jira integration. Use prowler_list_integrations with integration_type=['jira'] to find it."
         ),
-        project_key: str = Field(
+        project_key: NonBlankStr = Field(
             description="Key of the Jira project to read the issue types from (e.g. 'PROJ'). It must be one of the keys in the 'projects' mapping of the integration configuration."
         ),
     ) -> dict[str, Any]:
@@ -692,13 +664,13 @@ class IntegrationsTools(BaseTool):
 
     async def send_findings_to_jira(
         self,
-        integration_id: str = Field(
+        integration_id: NonBlankStr = Field(
             description="UUID of the Jira integration to send the findings through. It must be enabled."
         ),
-        project_key: str = Field(
+        project_key: NonBlankStr = Field(
             description="Key of the Jira project the work items are created in (e.g. 'PROJ'). It must be one of the keys in the 'projects' mapping of the integration configuration."
         ),
-        issue_type: str = Field(
+        issue_type: NonBlankStr = Field(
             description="Jira issue type for the created work items (e.g. 'Task', 'Bug', 'Story'). It must be one of the values returned by prowler_get_jira_issue_types for this project."
         ),
         finding_ids: list[str] = Field(
@@ -783,20 +755,26 @@ class IntegrationsTools(BaseTool):
                 return self._jira_dispatch_unknown(
                     task_id=None,
                     error=(
-                        f"the request that starts the dispatch failed on the server: {e} "
+                        "the request that starts the dispatch failed on Prowler's side. "
                         "It may have been queued anyway."
                     ),
                 )
 
             self.logger.error(f"Jira dispatch was rejected by Prowler: {e}")
             return self._jira_dispatch_rejected(str(e))
+        except CredentialError:
+            # Authentication happens before the request goes out, so nothing was
+            # queued. It is raised rather than reported as a dispatch outcome:
+            # there is no partial state to describe, and the shared classifier
+            # says what has to be fixed, which no retry of this call can.
+            raise
         except Exception as e:
             # No answer came back, so the request may still have been accepted
             self.logger.error(f"Jira dispatch could not be started: {e}")
             return self._jira_dispatch_unknown(
                 task_id=None,
                 error=(
-                    f"the request that starts the dispatch got no answer: {e} "
+                    "the request that starts the dispatch got no answer. "
                     "It may have been accepted anyway."
                 ),
             )
@@ -866,7 +844,7 @@ class IntegrationsTools(BaseTool):
         normalized = normalized.removesuffix(".atlassian.net")
 
         if not normalized:
-            raise ValueError(
+            raise InvalidArgument(
                 f"Invalid Jira domain: {domain}. Provide the Atlassian site name, for example "
                 "'acme' for the site 'https://acme.atlassian.net'."
             )
@@ -890,7 +868,7 @@ class IntegrationsTools(BaseTool):
             if not isinstance(credentials.get(key), str) or not credentials[key].strip()
         ]
         if missing:
-            raise ValueError(
+            raise InvalidArgument(
                 "Jira credentials are replaced as a whole, so 'domain', 'user_mail' and "
                 f"'api_token' are all required. Missing or empty: {', '.join(missing)}. "
                 "Sending an incomplete object would destroy the stored credentials and break "
@@ -908,29 +886,33 @@ class IntegrationsTools(BaseTool):
             try:
                 value = json.loads(value)
             except json.JSONDecodeError as e:
-                raise ValueError(f"Invalid JSON for {param_name}: {e}")
+                raise InvalidArgument(f"Invalid JSON for {param_name}: {e}") from e
 
         if not isinstance(value, dict):
-            raise ValueError(f"{param_name} must be a JSON object.")
+            raise InvalidArgument(f"{param_name} must be a JSON object.")
         return value
 
     async def _get_integration_raw(self, integration_id: str) -> dict[str, Any]:
         """Fetch the raw JSON:API resource of an integration.
 
         Raises:
-            ValueError: If the payload does not contain a usable integration resource
+            ToolError: If the payload does not contain a usable integration resource.
+                Raised without a ``from`` clause because these messages name the
+                integration and the tool that lists valid IDs, and the two cases
+                are reported differently: a missing resource is the caller's
+                mistake, a resource without attributes is the API's.
         """
         response = await self.api_client.get(f"/integrations/{integration_id}")
         integration = response.get("data")
 
         if not isinstance(integration, dict) or not integration.get("id"):
-            raise ValueError(
+            raise ToolError(
                 f"Integration {integration_id} was not found. Use prowler_list_integrations "
                 "to get a valid integration ID."
             )
 
         if not isinstance(integration.get("attributes"), dict):
-            raise ValueError(
+            raise ToolError(
                 f"Prowler returned integration {integration_id} without its attributes, so "
                 "its state cannot be read."
             )
@@ -970,7 +952,9 @@ class IntegrationsTools(BaseTool):
         integration_id = api_response.get("data", {}).get("id")
 
         if not integration_id:
-            raise ValueError(
+            # The integration may well exist, so this must not read as "nothing
+            # happened" and invite a duplicate.
+            raise ToolError(
                 "Prowler accepted the integration creation but did not return its ID, so the "
                 "connection could not be checked. Use prowler_list_integrations to see whether "
                 "the integration exists before creating it again."
@@ -981,11 +965,17 @@ class IntegrationsTools(BaseTool):
         try:
             integration = await self._get_integration_raw(integration_id)
         except Exception as e:
-            # The integration exists, so surface its ID instead of a plain read failure
+            # The integration exists, so surface its ID instead of a plain read
-            raise ValueError(
+            # failure. No `from` clause: a cause would let the shared classifier
-                f"Integration {integration_id} was created, but reading its state failed: {e} "
+            # replace this with a sentence that does not mention the ID. The
+            # failure text stays in the log, where the classifier would keep it.
+            self.logger.error(
+                f"Integration {integration_id} could not be read back: {e}"
+            )
+            raise ToolError(
+                f"Integration {integration_id} was created, but reading its state failed. "
                 "Use prowler_get_integration with that ID to check it."
-            ) from e
+            )
 
         return IntegrationConnectionStatus.create(
             integration, connection_status
@@ -1030,7 +1020,7 @@ class IntegrationsTools(BaseTool):
             return {
                 "connected": None,
                 "error": (
-                    f"The connection check could not be completed: {e} This says nothing "
+                    "The connection check could not be completed. This says nothing "
                     "about the stored credentials, run prowler_test_integration_connection "
                     "to check them again."
                 ),

mcp_server/prowler_mcp_server/prowler_app/tools/muting.py

@@ -8,8 +8,11 @@ This module provides tools for managing finding muting in Prowler, including:
 import json
 from typing import Any
 
+from fastmcp.exceptions import ToolError
 from pydantic import Field
 
+from prowler_mcp_server.lib.errors import InvalidArgument
+from prowler_mcp_server.lib.types import NonBlankStr
 from prowler_mcp_server.prowler_app.models.muting import (
     DetailedMuteRule,
     MutelistResponse,
@@ -28,10 +31,31 @@ class MutingTools(BaseTool):
 
     # ===== MUTELIST TOOLS =====
 
+    async def _get_mutelist_raw(self) -> dict[str, Any] | None:
+        """Return the tenant's mutelist, or None when it has none.
+
+        Returns:
+            The mutelist configuration, or None when the tenant has none
+        """
+        params = {
+            "filter[processor_type]": "mutelist",
+            "fields[processors]": "processor_type,configuration,inserted_at,updated_at",
+        }
+
+        clean_params = self.api_client.build_filter_params(params)
+        api_response = await self.api_client.get("/processors", params=clean_params)
+
+        data = api_response.get("data", [])
+        if not data:
+            return None
+
+        # Only one mutelist can exist per tenant
+        return MutelistResponse.from_api_response(data[0]).model_dump()
+
     async def get_mutelist(self) -> dict[str, Any]:
         """Retrieve the current mutelist configuration for the tenant.
 
-        IMPORTANT: Only one mutelist can exist per tenant. Returns an error message if no mutelist exists.
+        IMPORTANT: Only one mutelist can exist per tenant. Fails with a message saying so if no mutelist exists.
         For detailed information about mutelist structure and configuration, search Prowler documentation
         using prowler_docs_search tool available in this MCP Server.
 
@@ -47,26 +71,15 @@ class MutingTools(BaseTool):
         """
         self.logger.info("Retrieving mutelist configuration...")
 
-        # Query processors filtered by type=mutelist
+        mutelist = await self._get_mutelist_raw()
-        params = {
+        if mutelist is None:
-            "filter[processor_type]": "mutelist",
+            # No `from`: this names the tool that creates one, which the shared
-            "fields[processors]": "processor_type,configuration,inserted_at,updated_at",
+            # classifier cannot know.
-        }
+            raise ToolError(
-
+                "No mutelist configuration exists for this tenant. Use "
-        clean_params = self.api_client.build_filter_params(params)
+                "prowler_set_mutelist to create one."
-        api_response = await self.api_client.get("/processors", params=clean_params)
+            )
-
+        return mutelist
-        data = api_response.get("data", [])
-
-        if len(data) == 0:
-            return {
-                "error": "No mutelist found",
-                "message": "No mutelist configuration exists for this tenant. Use prowler_set_mutelist to create one.",
-            }
-
-        # Return the first (and only) mutelist
-        mutelist = MutelistResponse.from_api_response(data[0])
-        return mutelist.model_dump()
 
     async def set_mutelist(
         self,
@@ -128,9 +141,9 @@ Structure:
             configuration = json.loads(configuration)
 
         # Check if mutelist already exists
-        existing_mutelist = await self.get_mutelist()
+        existing_mutelist = await self._get_mutelist_raw()
 
-        if "error" in existing_mutelist:
+        if existing_mutelist is None:
             # Create new mutelist
             self.logger.info("Creating new mutelist...")
             create_body = {
@@ -183,21 +196,22 @@ Structure:
         self.logger.info("Deleting mutelist configuration...")
 
         # Get existing mutelist
-        existing_mutelist = await self.get_mutelist()
+        existing_mutelist = await self._get_mutelist_raw()
 
-        if "error" in existing_mutelist:
+        if existing_mutelist is None:
-            return {
+            raise ToolError(
-                "success": False,
+                "There is no mutelist configuration to delete. Use "
-                "message": "No mutelist found to delete",
+                "prowler_get_mutelist to confirm the current state."
-            }
+            )
 
         # Delete the mutelist
         mutelist_id = existing_mutelist["id"]
         await self.api_client.delete(f"/processors/{mutelist_id}")
 
+        # No success flag: a deletion that did not happen leaves this tool as an
+        # error, so there is no second shape for one to distinguish.
         return {
-            "success": True,
+            "message": "Mutelist deleted successfully. Findings it had muted stay muted."
-            "message": "Mutelist deleted successfully",
         }
 
     # ===== MUTE RULES TOOLS =====
@@ -268,7 +282,7 @@ Structure:
                 elif enabled.lower() == "false":
                     params["filter[enabled]"] = False
                 else:
-                    raise ValueError(
+                    raise InvalidArgument(
                         f"Invalid enabled value: {enabled}. Valid values are True, False, 'true', 'false' or None."
                     )
         if search:
@@ -282,7 +296,7 @@ Structure:
 
     async def get_mute_rule(
         self,
-        rule_id: str = Field(
+        rule_id: NonBlankStr = Field(
             description="UUID of the mute rule to retrieve. Must be a valid UUID format (e.g., '019ac0d6-90d5-73e9-9acf-c22e256f1bac')."
         ),
     ) -> dict[str, Any]:
@@ -316,10 +330,10 @@ Structure:
 
     async def create_mute_rule(
         self,
-        name: str = Field(
+        name: NonBlankStr = Field(
             description="Name for the mute rule. Should be descriptive and meaningful (e.g., 'Dev S3 Public Access', 'Test Environment IMDSv1')."
         ),
-        reason: str = Field(
+        reason: NonBlankStr = Field(
             description="Reason for muting these findings. Document why this security issue is acceptable or intentional (e.g., 'Development environment with controlled access', 'Legacy application requires IMDSv1')."
         ),
         finding_ids: list[str] = Field(
@@ -367,14 +381,14 @@ Structure:
 
     async def update_mute_rule(
         self,
-        rule_id: str = Field(
+        rule_id: NonBlankStr = Field(
             description="UUID of the mute rule to update. Must be a valid UUID format."
         ),
-        name: str | None = Field(
+        name: NonBlankStr | None = Field(
             default=None,
             description="New name for the rule. If not specified, name remains unchanged.",
         ),
-        reason: str | None = Field(
+        reason: NonBlankStr | None = Field(
             default=None,
             description="New reason for the rule. If not specified, reason remains unchanged.",
         ),
@@ -435,7 +449,7 @@ Structure:
 
     async def delete_mute_rule(
         self,
-        rule_id: str = Field(
+        rule_id: NonBlankStr = Field(
             description="UUID of the mute rule to delete. Must be a valid UUID format."
         ),
     ) -> dict[str, Any]:
@@ -457,15 +471,18 @@ Structure:
         """
         self.logger.info(f"Deleting mute rule {rule_id}...")
 
-        result = await self.api_client.delete(f"/mute-rules/{rule_id}")
+        # A deletion that did not happen answers with an error status, which
+        # leaves this tool as an error. Reaching this line means Prowler accepted
+        # it, whether it answered 204 with no body or 200 with the deleted
+        # resource, so there is no second outcome to report: the previous
+        # "Failed to delete mute rule" fired on the shape of the answer rather
+        # than on anything having gone wrong, and said nothing a caller could act
+        # on.
+        await self.api_client.delete(f"/mute-rules/{rule_id}")
 
-        if result.get("success"):
         return {
-                "success": True,
+            "message": (
-                "message": "Mute rule deleted successfully",
+                f"Mute rule {rule_id} deleted successfully. The findings it muted stay "
-            }
+                "muted."
-        else:
+            )
-            return {
-                "success": False,
-                "message": "Failed to delete mute rule",
         }

mcp_server/prowler_mcp_server/prowler_app/tools/providers.py

@@ -6,10 +6,14 @@ including searching, connecting, and deleting providers.
 
 from typing import Any
 
+from fastmcp.exceptions import ToolError
 from pydantic import Field
 
+from prowler_mcp_server.lib.errors import InvalidArgument
+from prowler_mcp_server.lib.types import NonBlankStr
 from prowler_mcp_server.prowler_app.models.providers import (
     ProviderConnectionStatus,
+    ProviderDeletionResult,
     ProvidersListResponse,
 )
 from prowler_mcp_server.prowler_app.tools.base import BaseTool
@@ -95,7 +99,7 @@ class ProvidersTools(BaseTool):
                 elif connected.lower() == "false":
                     params["filter[connected]"] = False
                 else:
-                    raise ValueError(
+                    raise InvalidArgument(
                         f"Invalid connected value: {connected}. Valid values are True, False, 'true', 'false' or None."
                     )
 
@@ -128,13 +132,13 @@ class ProvidersTools(BaseTool):
 
     async def connect_provider(
         self,
-        provider_uid: str = Field(
+        provider_uid: NonBlankStr = Field(
             description="Provider's unique identifier. For supported UID provider formats, please refer to Prowler Hub/Prowler Documentation that you can also find in form of tools in this MCP Server"
         ),
-        provider_type: str = Field(
+        provider_type: NonBlankStr = Field(
             description="Type of provider to be scanned with Prowler. Valid values include: 'aws', 'azure', 'gcp', 'kubernetes'... For more valid values, please refer to Prowler Hub/Prowler Documentation that you can also find in form of tools in this MCP Server."
         ),
-        alias: str | None = Field(
+        alias: NonBlankStr | None = Field(
             default=None,
             description="Human-friendly name for this provider. Optional but recommended for easy identification. Use descriptive names to distinguish multiple accounts of the same type.",
         ),
@@ -291,7 +295,7 @@ class ProvidersTools(BaseTool):
 
     async def delete_provider(
         self,
-        provider_id: str = Field(
+        provider_id: NonBlankStr = Field(
             description="Prowler's internal UUID (v4) for the provider to permanently remove, generated when the provider was registered in the system. Use `prowler_search_providers` tool to find the provider_id if you only know the alias or the provider's own identifier (provider_uid)"
         ),
     ) -> dict[str, Any]:
@@ -300,33 +304,120 @@ class ProvidersTools(BaseTool):
         WARNING: This is a destructive operation that cannot be undone. The provider will need to be
         re-added with prowler_connect_provider if you want to scan it again.
 
-        The tool always returns the deletion status and message.
+        Prowler removes the provider and everything attached to it (its scans, findings and
+        resources) in a background task, so a large provider can take longer than the time
+        this tool waits for it.
+
+        The result includes:
+        - status: 'deleted' when Prowler finished removing the provider, 'in_progress' when
+          the deletion was accepted and is still running
+        - task_id: the background task, present when the deletion was still running
+
+        NEVER send the deletion again while status='in_progress'. Use prowler_search_providers
+        to check whether the provider is gone.
         """
         self.logger.info(f"Deleting provider {provider_id}...")
-        try:
+
-            # Initiate the deletion task
+        # A failure of the request itself is left to the shared classifier: the
+        # deletion never started, so there is no partial state to describe.
         task_response = await self.api_client.delete(f"/providers/{provider_id}")
         task_id = task_response.get("data", {}).get("id")
 
-            # Poll until task completes (with 60 second timeout)
+        if not task_id:
+            # The deletion may well be running, so this must not read as "nothing
+            # happened". No `from` clause: this names the provider and the tool
+            # that checks it, neither of which the shared classifier can know.
+            raise ToolError(
+                f"Prowler accepted the deletion of provider {provider_id} but did not "
+                "return the ID of the background task, so its outcome cannot be checked. "
+                "Use prowler_search_providers to see whether the provider is still there "
+                "before sending the deletion again."
+            )
+
+        try:
             await self.api_client.poll_task_until_complete(
                 task_id=task_id, timeout=60, poll_interval=1.0
             )
-
-            # If we reach here, the task completed successfully
-            return {
-                "deleted": True,
-                "message": f"Provider {provider_id} deleted successfully",
-            }
         except Exception as e:
-            self.logger.error(f"Provider deletion failed: {e}")
+            self.logger.error(f"Provider deletion did not complete cleanly: {e}")
-            return {
+            return await self._provider_deletion_fallback(provider_id, task_id)
-                "deleted": False,
+
-                "message": f"Provider {provider_id} deletion failed: {str(e)}",
+        return ProviderDeletionResult(
-            }
+            status="deleted",
+            message=f"Provider {provider_id} deleted successfully",
+        ).model_dump()
 
     # Private helper methods
 
+    async def _provider_deletion_fallback(
+        self, provider_id: str, task_id: str
+    ) -> dict[str, Any]:
+        """Report a provider deletion whose polling did not end on a completed task.
+
+        Running out of the polling window is not a failure: Prowler removes the
+        provider together with its scans, findings and resources, which outlives
+        60 seconds on a large account. The deletion was accepted and is still
+        going, so calling it failed would be wrong twice over -- it is not, and
+        it invites a retry of a destructive call already in flight.
+
+        The task is read once more here, because polling gives up on the clock
+        rather than on the task: a deletion that finished just after the last
+        poll is a finished deletion and is reported as one.
+
+        Only a task that actually stopped is an error, and it is raised rather
+        than returned, because then the provider is still there.
+
+        Raises:
+            ToolError: If the deletion task ended without deleting the provider.
+                Raised without a ``from`` clause because the message names what
+                was left behind, which the shared classifier cannot know.
+        """
+        state = None
+        try:
+            task = await self.api_client.get(f"/tasks/{task_id}")
+            state = task.get("data", {}).get("attributes", {}).get("state")
+        except Exception as e:
+            self.logger.error(f"Could not read the state of task {task_id}: {e}")
+
+        if state == "completed":
+            # The deletion outran the polling window by a moment, not by more.
+            return ProviderDeletionResult(
+                status="deleted",
+                message=f"Provider {provider_id} deleted successfully",
+            ).model_dump()
+
+        if state in ("failed", "cancelled"):
+            # The failure that got us here is logged, not relayed: it carries
+            # upstream text, and the classifier masks exactly this kind of
+            # message when a tool does not write it itself.
+            raise ToolError(
+                f"The task deleting provider {provider_id} ended as '{state}', so the "
+                "provider was not deleted. Prowler removes a provider together with its "
+                "scans, findings and resources, so part of that may already be gone. Use "
+                "prowler_search_providers to check the current state."
+            )
+
+        if state is None:
+            message = (
+                f"The deletion of provider {provider_id} was accepted, but its progress "
+                "could not be read, so whether it finished is unknown. Do not "
+                "send the deletion again. Use prowler_search_providers to check whether "
+                "the provider is gone."
+            )
+        else:
+            message = (
+                f"The deletion of provider {provider_id} was accepted and is still "
+                f"running (task state '{state}'), which is normal for a provider with "
+                "many scans and findings. Do not send the deletion again. Use "
+                "prowler_search_providers to check whether it is gone."
+            )
+
+        return ProviderDeletionResult(
+            status="in_progress",
+            task_id=task_id,
+            message=message,
+        ).model_dump()
+
     async def _check_provider_exists(self, provider_uid: str) -> str | None:
         """Check if a provider already exists by its UID.
 
@@ -357,7 +448,7 @@ class ProvidersTools(BaseTool):
             return prowler_provider_id
         else:
             # Multiple providers with the same UID is a data integrity issue
-            raise Exception(
+            raise ToolError(
                 f"Data integrity error: Found {len(providers)} providers with UID '{provider_uid}'. "
                 f"Each provider UID should be unique. Please contact support or manually clean up duplicate providers."
             )
@@ -392,7 +483,11 @@ class ProvidersTools(BaseTool):
 
         provider_id = await self._check_provider_exists(provider_uid)
         if provider_id is None:
-            raise Exception(f"Provider {provider_uid} creation failed")
+            raise ToolError(
+                f"Prowler accepted the creation of provider {provider_uid} but the "
+                "provider cannot be found afterwards. Use prowler_search_providers to "
+                "check whether it exists before creating it again."
+            )
         return provider_id
 
     async def _update_provider_alias(
@@ -418,7 +513,10 @@ class ProvidersTools(BaseTool):
             f"/providers/{prowler_provider_id}", json_data=update_body
         )
         if result.get("data", {}).get("attributes", {}).get("alias") != alias:
-            raise Exception(f"Provider {prowler_provider_id} alias update failed")
+            raise ToolError(
+                f"Provider {prowler_provider_id} exists, but its alias was not updated. "
+                "Use prowler_search_providers to read its current alias."
+            )
 
     def _determine_secret_type(self, credentials: dict[str, Any]) -> str:
         """Determine the secret type from credentials structure.
@@ -443,9 +541,17 @@ class ProvidersTools(BaseTool):
             prowler_provider_id: The Prowler-generated provider ID
 
         Returns:
-            The secret ID if exists, None otherwise
+            The secret ID if the provider has one, None if it has none
+
+        Raises:
+            Exception: If the lookup itself failed, so that "no secret" is never
+                reported for a provider whose secret could not be read
         """
-        try:
+        # A failure here is not swallowed into None. None means "this provider has
+        # no secret", which sends `_store_credentials` down the create branch, and
+        # a provider holds at most one secret: creating a second one is refused,
+        # and the caller would be told its credentials were rejected when all that
+        # actually failed was this read.
         response = await self.api_client.get(
             "/providers/secrets",
             params={"filter[provider]": prowler_provider_id},
@@ -458,13 +564,8 @@ class ProvidersTools(BaseTool):
                 f"Found existing secret {secret_id} for provider {prowler_provider_id}"
             )
             return secret_id
-            else:
+
-                self.logger.info(
+        self.logger.info(f"No existing secret found for provider {prowler_provider_id}")
-                    f"No existing secret found for provider {prowler_provider_id}"
-                )
-                return None
-        except Exception as e:
-            self.logger.error(f"Error checking for existing secret: {e}")
         return None
 
     async def _get_secret_type(self, secret_id: str) -> str | None:
@@ -573,13 +674,24 @@ class ProvidersTools(BaseTool):
                 raise
 
     async def _test_connection(self, prowler_provider_id: str) -> dict[str, Any]:
-        """Test connection to a provider.
+        """Test connection to a provider and wait for the result.
+
+        A test that could not be run is reported as 'connected: None', which
+        `ProviderConnectionStatus` renders as 'not_tested', rather than as a
+        failure. Credentials that do not work come back as a completed task
+        carrying 'connected: False', so an exception here never describes them:
+        it means this server could not get the test run at all -- an expired
+        Prowler credential, a rate limit, a test that outlived the timeout.
+        Reporting that as 'failed' would blame the provider's credentials for
+        something they did not cause, and send the caller off to fix a working
+        role.
 
         Args:
             prowler_provider_id: The Prowler-generated provider ID
 
         Returns:
-            Connection status dictionary with 'connected' boolean and optional 'error' message
+            Connection status dictionary with a 'connected' boolean or None, and
+            an optional 'error' message
         """
         self.logger.info(f"Testing connection for provider {prowler_provider_id}...")
         try:
@@ -589,6 +701,11 @@ class ProvidersTools(BaseTool):
             )
             task_id = task_response.get("data", {}).get("id")
 
+            if not task_id:
+                raise ValueError(
+                    "Prowler did not return the ID of the connection test task."
+                )
+
             # Poll until task completes (with 60 second timeout)
             completed_task = await self.api_client.poll_task_until_complete(
                 task_id=task_id, timeout=60, poll_interval=1.0
@@ -596,13 +713,26 @@ class ProvidersTools(BaseTool):
 
             # Extract the result from the completed task
             task_result = (
-                completed_task.get("data", {}).get("attributes", {}).get("result", {})
+                completed_task.get("data", {}).get("attributes", {}).get("result")
+            )
+
+            if not isinstance(task_result, dict):
+                raise ValueError(
+                    "The connection test task completed without reporting a result."
                 )
 
             return task_result
         except Exception as e:
-            self.logger.error(f"Connection test failed: {e}")
+            self.logger.error(f"Connection test could not be completed: {e}")
-            return {"connected": False, "error": str(e)}
+            return {
+                "connected": None,
+                "error": (
+                    "The connection test could not be completed. This says nothing "
+                    "about the provider's credentials, they were never tested. Use "
+                    "prowler_search_providers to read the connection state Prowler has "
+                    "stored for this provider."
+                ),
+            }
 
     async def _get_final_provider_state(
         self, prowler_provider_id: str

mcp_server/prowler_mcp_server/prowler_app/tools/resources.py

@@ -8,6 +8,7 @@ from typing import Any
 
 from pydantic import Field
 
+from prowler_mcp_server.lib.types import NonBlankStr
 from prowler_mcp_server.prowler_app.models.resources import (
     DetailedResource,
     ResourceEventsResponse,
@@ -176,7 +177,7 @@ class ResourcesTools(BaseTool):
 
     async def get_resource(
         self,
-        resource_id: str = Field(
+        resource_id: NonBlankStr = Field(
             description="Prowler's internal UUID (v4) for the resource to retrieve, generated when the resource was discovered in the system. Use `prowler_list_resources` tool to find the right ID"
         ),
     ) -> dict[str, Any]:
@@ -347,7 +348,7 @@ class ResourcesTools(BaseTool):
 
     async def get_resource_events(
         self,
-        resource_id: str = Field(
+        resource_id: NonBlankStr = Field(
             description="Prowler's internal UUID (v4) for the resource. Use `prowler_list_resources` to find the right ID, or get it from a finding's resource relationship via `prowler_get_finding_details`."
         ),
         lookback_days: int = Field(

mcp_server/prowler_mcp_server/prowler_app/tools/roles.py

@@ -11,8 +11,11 @@ adding to it.
 
 from typing import Any
 
+from fastmcp.exceptions import ToolError
 from pydantic import Field
 
+from prowler_mcp_server.lib.errors import ProwlerAPIError
+from prowler_mcp_server.lib.types import NonBlankStr
 from prowler_mcp_server.prowler_app.models.roles import (
     DetailedRole,
     RolesListResponse,
@@ -70,7 +73,7 @@ class RolesTools(BaseTool):
 
     async def get_role(
         self,
-        role_id: str = Field(
+        role_id: NonBlankStr = Field(
             description="Prowler's internal UUID (v4) for the role to retrieve. Use `prowler_list_roles` to find role IDs if you only know a name."
         ),
     ) -> dict[str, Any]:
@@ -98,7 +101,7 @@ class RolesTools(BaseTool):
 
     async def get_user_roles(
         self,
-        user_id: str = Field(
+        user_id: NonBlankStr = Field(
             description="Prowler's internal UUID (v4) for the user whose roles you want. Use `prowler_list_users` to find user IDs, or `prowler_get_current_user` for the caller."
         ),
     ) -> dict[str, Any]:
@@ -124,10 +127,10 @@ class RolesTools(BaseTool):
 
     async def set_user_role(
         self,
-        user_id: str = Field(
+        user_id: NonBlankStr = Field(
             description="Prowler's internal UUID (v4) for the user whose role you want to set. Use `prowler_list_users` to find user IDs."
         ),
-        role_id: str = Field(
+        role_id: NonBlankStr = Field(
             description="Prowler's internal UUID (v4) for the role the user should hold. Use `prowler_list_roles` to find role IDs."
         ),
     ) -> dict[str, Any]:
@@ -166,11 +169,20 @@ class RolesTools(BaseTool):
         # user with no role at all. Confirm the role exists before replacing.
         try:
             await self.api_client.get(f"/roles/{role_id}")
-        except Exception as e:
+        except ProwlerAPIError as e:
-            raise ValueError(
+            if e.status_code != 404:
-                f"Role {role_id} could not be read ({e}), so user {user_id} was left "
+                # Only a not-found says anything about the role ID. A permission
-                f"unchanged. Use `prowler_list_roles` to find a valid role ID."
+                # error, a rate limit or a server error is about the request, so
-            ) from e
+                # it goes to the shared classifier rather than being reported as
+                # an ID the caller should replace.
+                raise
+            # No `from` clause: this says what state the user was left in, which
+            # the shared classifier cannot know, and a cause would let it replace
+            # this message with its own.
+            raise ToolError(
+                f"Role {role_id} does not exist in this tenant, so user {user_id} was "
+                f"left unchanged. Use `prowler_list_roles` to find a valid role ID."
+            )
 
         # PATCH replaces the user's whole role set with this single role, the
         # same call the Prowler UI makes when changing a user's role.

mcp_server/prowler_mcp_server/prowler_app/tools/scans.py

@@ -5,8 +5,10 @@ This module provides tools for managing and monitoring Prowler security scans.
 
 from typing import Any, Literal
 
+from fastmcp.exceptions import ToolError
 from pydantic import Field
 
+from prowler_mcp_server.lib.types import NonBlankStr
 from prowler_mcp_server.prowler_app.models.scans import (
     DetailedScan,
     ScanCreationResult,
@@ -127,7 +129,7 @@ class ScansTools(BaseTool):
 
     async def get_scan(
         self,
-        scan_id: str = Field(
+        scan_id: NonBlankStr = Field(
             description="Prowler's internal UUID (v4) for the scan to retrieve, generated when the scan was created (e.g., '123e4567-e89b-12d3-a456-426614174000'). Use `prowler_list_scans` tool to find scan IDs"
         ),
     ) -> dict[str, Any]:
@@ -171,10 +173,10 @@ class ScansTools(BaseTool):
 
     async def trigger_scan(
         self,
-        provider_id: str = Field(
+        provider_id: NonBlankStr = Field(
             description="Prowler's internal UUID (v4) for the provider to scan, generated when the provider was registered in the system (e.g., '4d0e2614-6385-4fa7-bf0b-c2e2f75c6877'). Use `prowler_search_providers` tool to find the provider ID"
         ),
-        name: str | None = Field(
+        name: NonBlankStr | None = Field(
             default=None,
             description="Optional human-friendly name for the scan. Use descriptive names to identify scan purpose or context, e.g., 'Weekly Production Security Audit', 'Pre-Deployment Validation', 'Compliance Check Q4 2025'",
         ),
@@ -191,7 +193,6 @@ class ScansTools(BaseTool):
         3. Use `prowler_get_scan` with the returned scan 'id' to monitor progress
         4. Once completed, use `prowler_search_security_findings` to analyze results
         """
-        try:
         # Build request data
         request_data: dict[str, Any] = {
             "data": {
@@ -222,29 +223,40 @@ class ScansTools(BaseTool):
         )
 
         if not scan_id:
-                raise Exception("No scan_id returned from scan creation")
+            # The scan may well have been queued, so this must not read as
+            # "nothing happened" and invite a duplicate run. No `from` clause:
+            # this names the provider and the tool that checks for the scan,
+            # neither of which the shared classifier can know.
+            raise ToolError(
+                "Prowler accepted the scan but did not return its ID, so it "
+                "cannot be looked up. Use prowler_list_scans for provider "
+                f"{provider_id} to see whether a scan is already running before "
+                "triggering another one."
+            )
 
-            self.logger.info(f"Scan created successfully: {scan_id}")
+        # The scan exists from here on, so a failure to read it back must name
+        # the ID rather than read as "the scan was not created".
+        try:
             scan_response = await self.api_client.get(f"/scans/{scan_id}")
             scan_info = DetailedScan.from_api_response(scan_response["data"])
+        except Exception as e:
+            # The failure itself is logged, not relayed: what it says is the
+            # shared classifier's to mask, and what the caller needs is the ID.
+            self.logger.error(f"Scan {scan_id} could not be read back: {e}")
+            raise ToolError(
+                f"Scan {scan_id} was created for provider {provider_id}, but reading "
+                "its state failed. Use prowler_get_scan with that ID to monitor "
+                "it. Do not trigger the scan again."
+            )
 
         return ScanCreationResult(
             scan=scan_info,
-                status="success",
             message=f"Scan {scan_id} created successfully. The scan may take some time to complete. Use prowler_get_scan tool with this ID to monitor progress.",
         ).model_dump()
 
-        except Exception as e:
-            self.logger.error(f"Scan creation failed: {e}")
-            return ScanCreationResult(
-                scan=None,
-                status="failed",
-                message=f"Scan creation failed: {str(e)}",
-            ).model_dump()
-
     async def schedule_daily_scan(
         self,
-        provider_id: str = Field(
+        provider_id: NonBlankStr = Field(
             description="Prowler's internal UUID (v4) for the provider to scan, generated when the provider was registered in the system (e.g., '4d0e2614-6385-4fa7-bf0b-c2e2f75c6877'). Use `prowler_search_providers` tool to find the provider ID"
         ),
     ) -> dict[str, Any]:
@@ -280,26 +292,49 @@ class ScansTools(BaseTool):
                 },
             },
         )
-        task_state = (
+
+        # Reaching this line means the schedule exists. Prowler commits the
+        # recurring schedule and its first scan inside the transaction that
+        # serves this request, so an answer at all means it was created; a
+        # provider that already has one is refused with a 409 instead, which
+        # leaves this tool as an error.
+        #
+        # The task in the answer is the FIRST scan run, queued to start a few
+        # seconds later, not the schedule. Its state therefore says nothing
+        # about whether the schedule was created, and reporting it as the
+        # outcome would call a schedule that exists a failure and invite a
+        # retry that can only hit that 409.
+        first_run_state = (
             task_response.get("data", {}).get("attributes", {}).get("state", None)
         )
 
-        if task_state == "available":
+        message = (
-            return_message = "Daily schedule created successfully. The schedule is being set up in the background. Use prowler_list_scans with provider_id filter to view scheduled scans."
+            f"Daily schedule created for provider {provider_id}. Prowler will scan it "
-        else:
+            "every 24 hours until the provider is deleted. Use prowler_list_scans with "
-            return_message = "Daily schedule creation failed. Please try again later."
+            "this provider_id and trigger='scheduled' to view its scheduled scans."
+        )
+
+        if first_run_state in ("failed", "cancelled"):
+            # Worth saying: the schedule stands, but the run that was supposed to
+            # start now will not produce findings, and only a manual scan fills
+            # the gap before tomorrow.
+            message = (
+                f"{message} Note that the first scan, which Prowler starts immediately, "
+                f"ended as '{first_run_state}'. The daily schedule is unaffected, but "
+                "use prowler_trigger_scan if you need results before the next run."
+            )
 
         return ScheduleCreationResult(
-            scheduled=(task_state == "available"),
+            first_run_state=first_run_state,
-            message=return_message,
+            message=message,
         ).model_dump()
 
     async def update_scan(
         self,
-        scan_id: str = Field(
+        scan_id: NonBlankStr = Field(
             description="Prowler's internal UUID (v4) for the scan to update, generated when the scan was created (e.g., '123e4567-e89b-12d3-a456-426614174000'). Use `prowler_list_scans` tool to find the scan ID if you only know the provider or scan name. Returns an error if the scan ID is invalid or not found."
         ),
-        name: str = Field(
+        name: NonBlankStr = Field(
             description="New human-friendly name for the scan (3-100 characters). Use descriptive names to improve organization and tracking, e.g., 'Production Security Audit - Q4 2025', 'Post-Deployment Compliance Check'. IMPORTANT: Only the scan name can be updated - other attributes (state, progress, duration) are read-only and managed by the system."
         ),
     ) -> dict[str, Any]:

mcp_server/prowler_mcp_server/prowler_app/tools/users.py

@@ -9,6 +9,7 @@ from typing import Any
 
 from pydantic import Field
 
+from prowler_mcp_server.lib.types import NonBlankStr
 from prowler_mcp_server.prowler_app.models.users import (
     DetailedUser,
     UsersListResponse,
@@ -79,7 +80,7 @@ class UsersTools(BaseTool):
 
     async def get_user(
         self,
-        user_id: str = Field(
+        user_id: NonBlankStr = Field(
             description="Prowler's internal UUID (v4) for the user to retrieve. Use `prowler_list_users` to find user IDs if you only know a name or email."
         ),
     ) -> dict[str, Any]:

mcp_server/prowler_mcp_server/prowler_app/utils/api_client.py

@@ -118,7 +118,21 @@ class ProwlerAPIClient(metaclass=SingletonMeta):
             if detail:
                 message = f"{message} - {detail}"
 
-            raise ProwlerAPIError(message, status, detail=detail) from e
+            # Carried on the exception, not into the message: a tool needs the
+            # body to tell an answer with an error status -- a 404 holding the
+            # empty result of a query that matched nothing -- apart from a
+            # request that actually failed.
+            try:
+                body = e.response.json()
+            except ValueError:
+                body = None
+
+            raise ProwlerAPIError(
+                message,
+                status,
+                detail=detail,
+                payload=body if isinstance(body, dict) else None,
+            ) from e
         except httpx.RequestError as e:
             # No answer came back, so whether the request was applied is unknown.
             logger.error(f"Error during {method.value} {path}: {e}")

mcp_server/prowler_mcp_server/prowler_app/utils/auth.py

@@ -6,6 +6,7 @@ from datetime import datetime
 from fastmcp.server.dependencies import get_http_headers
 
 from prowler_mcp_server import __version__
+from prowler_mcp_server.lib.errors import CredentialError
 from prowler_mcp_server.lib.logger import logger
 
 
@@ -64,7 +65,12 @@ class ProwlerAppAuth:
 
             # Decode and parse JSON
             decoded = base64.b64decode(base64_payload).decode("utf-8")
-            return json.loads(decoded)
+            payload = json.loads(decoded)
+
+            # A JWT payload is a JSON object. A list or a scalar decodes just as
+            # cleanly, so the type is checked here rather than left to blow up as
+            # an AttributeError on the first claim read.
+            return payload if isinstance(payload, dict) else None
         except Exception as e:
             logger.warning(f"Failed to parse JWT token: {e}")
             return None
@@ -76,14 +82,16 @@ class ProwlerAppAuth:
             authorization_header = headers.get("authorization", None)
 
             if not authorization_header:
-                raise ValueError("No authorization header provided")
+                raise CredentialError("No Authorization header was sent")
 
-            # Extract token from Bearer header
+            # Extract token from Bearer header. Authentication scheme names are
-            if authorization_header.startswith("Bearer "):
+            # case-insensitive (RFC 7235), and only the scheme prefix is removed:
-                token = authorization_header.replace("Bearer ", "")
+            # a token that happens to contain the word again keeps it.
-            else:
+            scheme, _, credential = authorization_header.partition(" ")
-                raise ValueError(
+            token = credential.strip()
-                    "Invalid authorization header format. Expected 'Bearer <token>'"
+            if scheme.lower() != "bearer" or not token:
+                raise CredentialError(
+                    "The Authorization header is not in 'Bearer <token>' form"
                 )
 
             # Check if it's an API key or JWT token
@@ -94,17 +102,29 @@ class ProwlerAppAuth:
                 # JWT token - validate and check expiration
                 payload = self._parse_jwt(token)
                 if not payload:
-                    raise ValueError("Invalid JWT token format")
+                    raise CredentialError("The token is not a readable JWT")
+
+                # Check if token is expired. `exp` is a numeric date in the
+                # spec, so a missing or non-numeric one makes the token
+                # unusable rather than merely stale -- comparing it would raise
+                # a TypeError and leave the failure masked as unclassified.
+                exp = payload.get("exp")
+                if isinstance(exp, bool) or not isinstance(exp, (int, float)):
+                    raise CredentialError(
+                        "The token carries no readable 'exp' expiration claim"
+                    )
 
-                # Check if token is expired
                 now = int(datetime.now().timestamp())
-                exp = payload.get("exp", 0)
                 if exp <= now:
-                    raise ValueError("Token has expired")
+                    raise CredentialError("The token has expired")
 
                 return token
         else:
-            raise ValueError(f"Invalid mode: {self.mode}")
+            # PROWLER_MCP_TRANSPORT_MODE holds something this server does not
+            # support. Nothing about a call caused it and nothing about a call
+            # can fix it, so it stays unclassified: masked for the model, logged
+            # for whoever runs the server.
+            raise RuntimeError(f"Invalid mode: {self.mode}")
 
     async def get_valid_token(self) -> str:
         """Get a valid token (API key or JWT token)."""

mcp_server/prowler_mcp_server/prowler_documentation/server.py

@@ -3,6 +3,7 @@ from typing import Any
 from fastmcp import FastMCP
 from pydantic import Field
 
+from prowler_mcp_server.lib.types import NonBlankStr
 from prowler_mcp_server.prowler_documentation.search_engine import (
     ProwlerDocsSearchEngine,
 )
@@ -14,7 +15,9 @@ prowler_docs_search_engine = ProwlerDocsSearchEngine()
 
 @docs_mcp_server.tool()
 def search(
-    term: str = Field(description="The term to search for in the documentation"),
+    term: NonBlankStr = Field(
+        description="The term to search for in the documentation"
+    ),
     page_size: int = Field(
         5,
         description="Number of top results to return. It must be between 1 and 20.",
@@ -39,7 +42,7 @@ def search(
 
 @docs_mcp_server.tool()
 def get_document(
-    doc_path: str = Field(
+    doc_path: NonBlankStr = Field(
         description="Path to the documentation file to retrieve. It is the same as the 'path' field of the search results. Use `prowler_docs_search` to find the path first."
     ),
 ) -> dict[str, str]:

mcp_server/prowler_mcp_server/prowler_hub/server.py

@@ -9,6 +9,7 @@ from fastmcp import FastMCP
 from pydantic import Field
 
 from prowler_mcp_server import __version__
+from prowler_mcp_server.lib.types import NonBlankStr
 
 # Initialize FastMCP for Prowler Hub
 hub_mcp_server = FastMCP("prowler-hub")
@@ -149,7 +150,7 @@ async def list_checks(
 
 @hub_mcp_server.tool()
 async def semantic_search_checks(
-    term: str = Field(
+    term: NonBlankStr = Field(
         description="Search term. Examples: 'public access', 'encryption', 'MFA', 'logging'.",
     ),
 ) -> dict:
@@ -208,7 +209,7 @@ async def semantic_search_checks(
 
 @hub_mcp_server.tool()
 async def get_check_details(
-    check_id: str = Field(
+    check_id: NonBlankStr = Field(
         description="The check ID to retrieve details for. Example: 's3_bucket_level_public_access_block'"
     ),
 ) -> dict:
@@ -346,10 +347,10 @@ async def get_check_details(
 
 @hub_mcp_server.tool()
 async def get_check_code(
-    provider_id: str = Field(
+    provider_id: NonBlankStr = Field(
         description="Prowler Provider ID. Example: 'aws', 'azure', 'gcp', 'kubernetes'. Use `prowler_hub_list_providers` to get available provider IDs.",
     ),
-    check_id: str = Field(
+    check_id: NonBlankStr = Field(
         description="The check ID. Example: 's3_bucket_public_access'. Get IDs from `prowler_hub_list_checks` or `prowler_hub_search_checks`.",
     ),
 ) -> dict:
@@ -392,10 +393,10 @@ async def get_check_code(
 
 @hub_mcp_server.tool()
 async def get_check_fixer(
-    provider_id: str = Field(
+    provider_id: NonBlankStr = Field(
         description="Prowler Provider ID. Example: 'aws', 'azure', 'gcp', 'kubernetes'. Use `prowler_hub_list_providers` to get available provider IDs.",
     ),
-    check_id: str = Field(
+    check_id: NonBlankStr = Field(
         description="The check ID. Example: 's3_bucket_public_access'. Get IDs from `prowler_hub_list_checks` or `prowler_hub_search_checks`.",
     ),
 ) -> dict:
@@ -517,7 +518,7 @@ async def list_compliances(
 
 @hub_mcp_server.tool()
 async def semantic_search_compliances(
-    term: str = Field(
+    term: NonBlankStr = Field(
         description="Search term. Examples: 'CIS', 'HIPAA', 'PCI', 'GDPR', 'SOC2', 'NIST'.",
     ),
 ) -> dict:
@@ -568,7 +569,7 @@ async def semantic_search_compliances(
 
 @hub_mcp_server.tool()
 async def get_compliance_details(
-    compliance_id: str = Field(
+    compliance_id: NonBlankStr = Field(
         description="The compliance framework ID to retrieve details for. Example: 'cis_4.0_aws'. Use `prowler_hub_list_compliances` or `prowler_hub_semantic_search_compliances` to find available compliance IDs.",
     ),
 ) -> dict:
@@ -708,7 +709,7 @@ async def list_providers() -> dict:
 
 @hub_mcp_server.tool()
 async def get_provider_services(
-    provider_id: str = Field(
+    provider_id: NonBlankStr = Field(
         description="The provider ID to get services for. Example: 'aws', 'azure', 'gcp', 'kubernetes'. Use `prowler_hub_list_providers` to get available provider IDs.",
     ),
 ) -> dict:

mcp_server/tests/lib/test_errors.py

@@ -11,7 +11,11 @@ import pytest
 from fastmcp import Client
 from pydantic import BaseModel, ValidationError
 
-from prowler_mcp_server.lib.errors import InvalidArgument, _describe_failure
+from prowler_mcp_server.lib.errors import (
+    CredentialError,
+    InvalidArgument,
+    _describe_failure,
+)
 from prowler_mcp_server.prowler_app.utils.api_client import (
     ProwlerAPIError,
     ProwlerAPIInvalidResponse,
@@ -99,6 +103,19 @@ def test_an_argument_this_server_rejected_is_repeated_verbatim():
     assert message == "page_size must be between 1 and 1000."
 
 
+def test_a_credential_caught_here_is_answered_like_the_401_it_would_have_got():
+    """It is not an argument problem, and saying so stops a pointless retry."""
+    message = _describe_failure(CredentialError("the token has expired"))
+
+    assert "the token has expired" in message
+    assert "changing the arguments will not help" in message
+
+
+def test_a_transport_this_server_cannot_serve_is_left_masked():
+    """No call caused a bad PROWLER_MCP_TRANSPORT_MODE and no call can fix it."""
+    assert _describe_failure(RuntimeError("Invalid mode: websocket")) is None
+
+
 def test_a_pydantic_rejection_names_the_field_without_echoing_the_value():
     """Pydantic quotes `input_value` back, and these tools take credentials."""
 
@@ -195,3 +212,18 @@ async def test_an_unreadable_api_answer_does_not_reach_the_agent_as_a_bad_argume
     assert result.isError is True
     assert "gateway timeout" not in result.content[0].text
     assert "argument" not in result.content[0].text
+
+
+async def test_a_tool_specific_message_survives_masking(
+    mcp_root_server, mock_api_client, mock_router
+):
+    """A `ToolError` raised without a `from` clause is the final word."""
+    mock_router.add("GET", "/api/v1/integrations/i1", json={"data": None})
+
+    async with Client(mcp_root_server) as client:
+        result = await client.call_tool_mcp(
+            "prowler_get_integration", {"integration_id": "i1"}
+        )
+
+    assert result.isError is True
+    assert "prowler_list_integrations" in result.content[0].text

mcp_server/tests/lib/test_types.py

@@ -0,0 +1,125 @@
+"""Tests for the argument types every tool shares.
+
+The bug these pin: "required" alone does not stop a blank identifier. A model
+that has no scan or query id to hand sends ``""`` rather than omitting the
+argument, and an unguarded empty string travels into a URL path or a request
+body -- where it comes back as a 404, or as an API rejection ("This field may
+not be blank") that names no argument and leaves the model with nothing to fix.
+"""
+
+import pytest
+from fastmcp import Client
+
+from tests.helpers.jsonapi import jsonapi_collection, jsonapi_resource
+
+SCAN_ID = "019ac0d6-90d5-73e9-9acf-c22e256f1bac"
+QUERIES = f"/api/v1/attack-paths-scans/{SCAN_ID}/queries"
+
+
+@pytest.mark.parametrize("query_id", ["", "   "], ids=["empty", "whitespace-only"])
+async def test_a_blank_identifier_is_rejected_before_any_request_goes_out(
+    mcp_root_server, mock_api_client, mock_router, query_id
+):
+    """The reported failure: a blank `query_id` reached Prowler as a 400.
+
+    The message has to name the argument. Prowler's own answer to the blank value
+    ("This field may not be blank") does not say which field, so the model had no
+    way to tell `scan_id` from `query_id` from the reply.
+    """
+    async with Client(mcp_root_server) as client:
+        result = await client.call_tool_mcp(
+            "prowler_run_attack_paths_query",
+            {"scan_id": SCAN_ID, "query_id": query_id},
+        )
+
+    assert result.isError is True
+    assert "query_id" in result.content[0].text
+    assert mock_router.requests == []
+
+
+async def test_an_identifier_keeps_its_surrounding_whitespace_out_of_the_url(
+    mcp_root_server, mock_api_client, mock_router
+):
+    """A padded id is the same id, and a raw one would build a URL-escaped path."""
+    mock_router.add("GET", QUERIES, json=jsonapi_collection([]))
+
+    async with Client(mcp_root_server) as client:
+        result = await client.call_tool_mcp(
+            "prowler_list_attack_paths_queries", {"scan_id": f"  {SCAN_ID}  "}
+        )
+
+    assert result.isError is False
+    assert mock_router.paths() == [f"GET {QUERIES}"]
+
+
+async def test_a_blank_optional_value_is_rejected_rather_than_written(
+    mcp_root_server, mock_api_client, mock_router
+):
+    """An omitted optional means "leave it alone"; a blank one would blank the field.
+
+    The API refuses it, so the only difference an unguarded blank makes is a
+    round trip and an error that names nothing.
+    """
+    async with Client(mcp_root_server) as client:
+        result = await client.call_tool_mcp(
+            "prowler_update_mute_rule", {"rule_id": SCAN_ID, "name": ""}
+        )
+
+    assert result.isError is True
+    assert "name" in result.content[0].text
+    assert mock_router.requests == []
+
+
+async def test_an_omitted_optional_string_is_still_omitted(
+    mcp_root_server, mock_api_client, mock_router
+):
+    """`NonBlankStr | None` must not turn "not provided" into a rejection."""
+    mock_router.add(
+        "GET",
+        f"/api/v1/mute-rules/{SCAN_ID}",
+        json={
+            "data": jsonapi_resource(
+                "mute-rules",
+                SCAN_ID,
+                {
+                    "name": "unchanged",
+                    "reason": "already reviewed",
+                    "enabled": True,
+                    "finding_uids": [],
+                },
+            )
+        },
+    )
+
+    async with Client(mcp_root_server) as client:
+        result = await client.call_tool_mcp(
+            "prowler_update_mute_rule", {"rule_id": SCAN_ID}
+        )
+
+    assert result.isError is False
+
+
+async def test_every_required_string_argument_is_guarded_against_a_blank(
+    mcp_root_server,
+):
+    """A guard only one tool carries is one the next tool will be written without.
+
+    Declared as `minLength` rather than checked inside each tool, so a client sees
+    the constraint in the schema before it calls.
+    """
+    async with Client(mcp_root_server) as client:
+        tools = await client.list_tools()
+
+    unguarded = [
+        f"{tool.name}.{name}"
+        for tool in tools
+        for name, schema in tool.inputSchema.get("properties", {}).items()
+        # Plain required strings only. A union such as `dict | str` takes a JSON
+        # string, where a blank is a parse failure the classifier already
+        # explains, and a blank filter is a filter that matches everything.
+        if schema.get("type") == "string"
+        and name in tool.inputSchema.get("required", [])
+        and schema.get("minLength") != 1
+    ]
+
+    assert unguarded == []

mcp_server/tests/prowler_app/tools/test_attack_paths.py

@@ -0,0 +1,225 @@
+"""Tests for the Attack Paths tools.
+
+An Attack Paths scan is a separate resource from a regular scan, with IDs of its
+own, and Prowler only creates one for an AWS provider. So the 404 these tools get
+is almost always a regular scan ID passed where an Attack Paths one belongs --
+and Prowler's own reason for it, a bare "Not found.", names neither the resource
+it looked in nor the tool that returns the right ID.
+"""
+
+import pytest
+from fastmcp import Client
+
+from tests.helpers.jsonapi import jsonapi_collection, jsonapi_resource
+
+QUERIES = "/api/v1/attack-paths-scans/s1/queries"
+
+
+async def test_an_id_that_is_not_an_attack_paths_scan_says_which_tool_returns_one(
+    mcp_root_server, mock_api_client, mock_router
+):
+    """Relaying "Not found." sends an agent to re-check an ID it cannot fix.
+
+    The reply has to name the confusion it stands for: regular scan IDs do not
+    resolve here, and only AWS providers have an Attack Paths scan at all.
+    """
+    mock_router.add(
+        "GET",
+        QUERIES,
+        status=404,
+        json={"errors": [{"status": "404", "detail": "Not found."}]},
+    )
+
+    async with Client(mcp_root_server) as client:
+        with pytest.raises(Exception, match="different resource from regular scans"):
+            await client.call_tool(
+                "prowler_list_attack_paths_queries", {"scan_id": "s1"}
+            )
+
+
+async def test_the_answer_names_the_tool_that_returns_a_usable_id(
+    mcp_root_server, mock_api_client, mock_router
+):
+    """An explanation with no next step leaves the agent guessing IDs."""
+    mock_router.add(
+        "GET",
+        QUERIES,
+        status=404,
+        json={"errors": [{"status": "404", "detail": "Not found."}]},
+    )
+
+    async with Client(mcp_root_server) as client:
+        with pytest.raises(Exception, match="prowler_list_attack_paths_scans"):
+            await client.call_tool(
+                "prowler_list_attack_paths_queries", {"scan_id": "s1"}
+            )
+
+
+async def test_a_failure_that_is_not_a_404_keeps_the_shared_message(
+    mcp_root_server, mock_api_client, mock_router
+):
+    """Only the 404 means a bad ID. A 403 is a permission the ID cannot fix."""
+    mock_router.add(
+        "GET",
+        QUERIES,
+        status=403,
+        json={"errors": [{"status": "403", "detail": "Denied."}]},
+    )
+
+    async with Client(mcp_root_server) as client:
+        with pytest.raises(Exception, match="prowler_get_current_user"):
+            await client.call_tool(
+                "prowler_list_attack_paths_queries", {"scan_id": "s1"}
+            )
+
+
+async def test_queries_come_back_as_a_list(
+    mcp_root_server, mock_api_client, mock_router
+):
+    """The success path is unchanged."""
+    mock_router.add(
+        "GET",
+        QUERIES,
+        json=jsonapi_collection(
+            [
+                jsonapi_resource(
+                    "attack-paths-queries",
+                    "aws-ec2-instances-internet-exposed",
+                    {
+                        "name": "Internet exposed EC2",
+                        "description": "Find internet-exposed EC2 instances",
+                        "provider": "aws",
+                        "parameters": [],
+                    },
+                )
+            ]
+        ),
+    )
+
+    async with Client(mcp_root_server) as client:
+        result = await client.call_tool(
+            "prowler_list_attack_paths_queries", {"scan_id": "s1"}
+        )
+
+    assert result.data[0]["id"] == "aws-ec2-instances-internet-exposed"
+
+
+# ------------------------------------------------------------- running a query
+
+RUN = "/api/v1/attack-paths-scans/s1/queries/run"
+SCHEMA = "/api/v1/attack-paths-scans/s1/schema"
+
+EMPTY_RESULT = {
+    "data": {
+        "type": "attack-paths-query-results",
+        "id": "s1",
+        "attributes": {"nodes": [], "relationships": []},
+    }
+}
+
+
+def _run_args(query_id: str = "aws-ec2-instances-internet-exposed") -> dict[str, str]:
+    """Arguments for a query run against the mocked scan."""
+    return {"scan_id": "s1", "query_id": query_id}
+
+
+async def test_a_query_that_matched_nothing_is_an_answer_not_a_failure(
+    mcp_root_server, mock_api_client, mock_router
+):
+    """Prowler answers a query that matched nothing with 404 and the result body.
+
+    Raising on the status called a clean account a failed call and sent the agent
+    off to re-check arguments that were right.
+    """
+    mock_router.add("POST", RUN, status=404, json=EMPTY_RESULT)
+
+    async with Client(mcp_root_server) as client:
+        result = await client.call_tool_mcp(
+            "prowler_run_attack_paths_query", _run_args()
+        )
+
+    assert result.isError is False
+    assert "matched nothing" in result.structuredContent["message"]
+
+
+async def test_an_empty_result_does_not_come_back_as_an_empty_object(
+    mcp_root_server, mock_api_client, mock_router
+):
+    """The serializer drops empty lists, so `{}` is all that would be left."""
+    mock_router.add("POST", RUN, status=404, json=EMPTY_RESULT)
+
+    async with Client(mcp_root_server) as client:
+        result = await client.call_tool("prowler_run_attack_paths_query", _run_args())
+
+    assert result.data != {}
+
+
+async def test_a_null_graph_is_read_as_an_empty_one(
+    mcp_root_server, mock_api_client, mock_router
+):
+    """Prowler can spell the empty graph as `null` rather than as empty lists.
+
+    Reading `null` as if it were a graph crashed the parse, turning the same
+    "nothing matched" answer into an error the agent could not act on.
+    """
+    mock_router.add("POST", RUN, status=404, json={"data": {"attributes": None}})
+
+    async with Client(mcp_root_server) as client:
+        result = await client.call_tool_mcp(
+            "prowler_run_attack_paths_query", _run_args()
+        )
+
+    assert result.isError is False
+    assert "matched nothing" in result.structuredContent["message"]
+
+
+async def test_a_run_against_an_unknown_scan_still_names_the_confusion(
+    mcp_root_server, mock_api_client, mock_router
+):
+    """A 404 with no result body is the ID being wrong, not an empty answer."""
+    mock_router.add(
+        "POST",
+        RUN,
+        status=404,
+        json={"errors": [{"status": "404", "detail": "Not found."}]},
+    )
+
+    async with Client(mcp_root_server) as client:
+        with pytest.raises(Exception, match="different resource from regular scans"):
+            await client.call_tool("prowler_run_attack_paths_query", _run_args())
+
+
+async def test_a_scan_whose_graph_records_no_schema_says_so(
+    mcp_root_server, mock_api_client, mock_router
+):
+    """This 404 is about the graph, not the ID, so it must not blame the ID."""
+    mock_router.add(
+        "GET",
+        SCHEMA,
+        status=404,
+        json={"detail": "No cartography schema metadata found for this provider"},
+    )
+
+    async with Client(mcp_root_server) as client:
+        with pytest.raises(Exception, match="no Cartography schema recorded"):
+            await client.call_tool(
+                "prowler_get_attack_paths_cartography_schema", {"scan_id": "s1"}
+            )
+
+
+async def test_a_schema_request_for_an_unknown_scan_names_the_confusion(
+    mcp_root_server, mock_api_client, mock_router
+):
+    """The other 404 here is the ID, and Prowler writes a JSON:API error for it."""
+    mock_router.add(
+        "GET",
+        SCHEMA,
+        status=404,
+        json={"errors": [{"status": "404", "detail": "Not found."}]},
+    )
+
+    async with Client(mcp_root_server) as client:
+        with pytest.raises(Exception, match="different resource from regular scans"):
+            await client.call_tool(
+                "prowler_get_attack_paths_cartography_schema", {"scan_id": "s1"}
+            )

mcp_server/tests/prowler_app/tools/test_compliance.py

@@ -0,0 +1,102 @@
+"""Tests for the compliance tools.
+
+Both compliance tools answer for exactly one scan. ``scan_id`` names it
+directly; ``provider_id`` names it indirectly, as "the latest completed scan of
+this provider". Passing both is not a refinement of either -- the scan the
+caller named may belong to a different provider entirely -- so it is rejected
+rather than resolved by preferring one, which would answer confidently for a
+provider nobody asked about.
+
+Tools are driven through an in-memory MCP client so FastMCP resolves the
+pydantic ``Field`` defaults and a raised failure arrives the way a client sees
+it.
+"""
+
+import pytest
+from fastmcp import Client
+
+from tests.helpers.jsonapi import jsonapi_collection, jsonapi_resource
+
+SCANS = "/api/v1/scans"
+OVERVIEWS = "/api/v1/compliance-overviews"
+REQUIREMENTS = f"{OVERVIEWS}/requirements"
+
+TOOLS = [
+    "prowler_get_compliance_overview",
+    "prowler_get_compliance_framework_state_details",
+]
+
+
+def arguments(tool: str, **overrides) -> dict:
+    """Build the arguments for either tool, which differ only in compliance_id."""
+    payload = dict(overrides)
+    if tool.endswith("framework_state_details"):
+        payload["compliance_id"] = "cis_1.5_aws"
+    return payload
+
+
+@pytest.mark.parametrize("tool", TOOLS)
+async def test_neither_a_scan_nor_a_provider_is_refused_before_any_request(
+    mcp_root_server, mock_api_client, mock_router, tool
+):
+    """There is no scan to answer for, and no way to guess one."""
+    async with Client(mcp_root_server) as client:
+        with pytest.raises(Exception, match="must be provided"):
+            await client.call_tool(tool, arguments(tool))
+
+    assert mock_router.paths() == []
+
+
+@pytest.mark.parametrize("tool", TOOLS)
+async def test_a_scan_and_a_provider_together_are_refused_rather_than_reconciled(
+    mcp_root_server, mock_api_client, mock_router, tool
+):
+    """Silently keeping the scan would answer for whichever provider owns it.
+
+    That report names a scan the caller did ask for, so nothing about it looks
+    wrong -- while the provider they also named went unread.
+    """
+    async with Client(mcp_root_server) as client:
+        with pytest.raises(Exception, match="not both"):
+            await client.call_tool(
+                tool, arguments(tool, scan_id="s1", provider_id="p1")
+            )
+
+    assert mock_router.paths() == []
+
+
+@pytest.mark.parametrize("tool", TOOLS)
+async def test_a_provider_on_its_own_resolves_to_its_latest_completed_scan(
+    mcp_root_server, mock_api_client, mock_router, tool
+):
+    """The indirection is the point of accepting a provider at all."""
+    mock_router.add(
+        "GET",
+        SCANS,
+        json=jsonapi_collection(
+            [jsonapi_resource("scans", "s9", {"state": "completed"})]
+        ),
+    )
+    mock_router.add("GET", OVERVIEWS, json=jsonapi_collection([]))
+    mock_router.add("GET", REQUIREMENTS, json=jsonapi_collection([]))
+    # Each tool reads the compliance state from its own endpoint; both filter it
+    # by the scan that had to be resolved first.
+    read = OVERVIEWS if tool.endswith("overview") else REQUIREMENTS
+
+    async with Client(mcp_root_server) as client:
+        await client.call_tool(tool, arguments(tool, provider_id="p1"))
+
+    assert mock_router.query_params("GET", SCANS)["filter[provider]"] == "p1"
+    assert mock_router.query_params("GET", read)["filter[scan_id]"] == "s9"
+
+
+@pytest.mark.parametrize("tool", TOOLS)
+async def test_a_provider_with_no_completed_scan_is_named_as_the_bad_argument(
+    mcp_root_server, mock_api_client, mock_router, tool
+):
+    """Nothing has been scanned yet, so there is no compliance state to report."""
+    mock_router.add("GET", SCANS, json=jsonapi_collection([]))
+
+    async with Client(mcp_root_server) as client:
+        with pytest.raises(Exception, match="No completed scans found for provider p1"):
+            await client.call_tool(tool, arguments(tool, provider_id="p1"))

mcp_server/tests/prowler_app/tools/test_integrations.py

@@ -312,7 +312,8 @@ async def test_creating_a_jira_integration_rejects_an_empty_domain(
 ):
     """A domain that normalizes to nothing is caught before the round trip."""
     async with Client(mcp_root_server) as client:
-        result = await client.call_tool(
+        with pytest.raises(Exception, match="Invalid Jira domain"):
+            await client.call_tool(
                 "prowler_create_jira_integration",
                 {
                     "domain": "https://",
@@ -321,8 +322,6 @@ async def test_creating_a_jira_integration_rejects_an_empty_domain(
                 },
             )
 
-    assert result.data["status"] == "failed"
-    assert "Invalid Jira domain" in result.data["error"]
     assert mock_router.requests == []
 
 
@@ -334,14 +333,10 @@ async def test_creating_a_jira_integration_rejects_an_empty_domain(
     ],
     ids=["security-hub", "amazon-s3"],
 )
-async def test_a_rejected_creation_is_reported_rather_than_raised(
+async def test_a_rejected_creation_fails_with_the_api_reason(
     mcp_root_server, mock_api_client, mock_router, tool, arguments
 ):
-    """Write tools answer with an error object so the agent can act on it.
+    """A refused creation is a tool error, and it still carries the API's reason."""
-
-    A raised exception reaches the model as a tool failure with no detail, and
-    the API's message is exactly what tells it what to do next.
-    """
     mock_router.add(
         "POST",
         INTEGRATIONS,
@@ -350,10 +345,8 @@ async def test_a_rejected_creation_is_reported_rather_than_raised(
     )
 
     async with Client(mcp_root_server) as client:
-        result = await client.call_tool(tool, arguments)
+        with pytest.raises(Exception, match="already has this integration"):
-
+            await client.call_tool(tool, arguments)
-    assert result.data["status"] == "failed"
-    assert "already has this integration" in result.data["error"]
 
 
 async def test_a_creation_with_no_id_back_warns_before_a_blind_retry(
@@ -367,12 +360,11 @@ async def test_a_creation_with_no_id_back_warns_before_a_blind_retry(
     mock_router.add("POST", INTEGRATIONS, json={"data": {}})
 
     async with Client(mcp_root_server) as client:
-        result = await client.call_tool(
+        with pytest.raises(Exception, match="did not return its ID"):
+            await client.call_tool(
                 "prowler_create_amazon_s3_integration", {"bucket_name": "my-reports"}
             )
 
-    assert result.data["status"] == "failed"
-    assert "did not return its ID" in result.data["error"]
     assert mock_router.paths() == [f"POST {INTEGRATIONS}"]
 
 
@@ -395,13 +387,11 @@ async def test_a_creation_whose_read_back_fails_still_hands_over_the_id(
     )
 
     async with Client(mcp_root_server) as client:
-        result = await client.call_tool(
+        with pytest.raises(Exception, match="Integration i1 was created"):
+            await client.call_tool(
                 "prowler_create_amazon_s3_integration", {"bucket_name": "my-reports"}
             )
 
-    assert result.data["status"] == "failed"
-    assert "Integration i1 was created" in result.data["error"]
-
 
 async def test_a_connection_check_that_cannot_run_is_not_reported_as_a_failure(
     mcp_root_server, mock_api_client, mock_router
@@ -542,13 +532,12 @@ async def test_a_configuration_that_is_not_an_object_is_rejected_before_the_writ
     stub_integration(mock_router, S3_ATTRIBUTES)
 
     async with Client(mcp_root_server) as client:
-        result = await client.call_tool(
+        with pytest.raises(Exception, match=message):
+            await client.call_tool(
                 "prowler_update_integration",
                 {"integration_id": "i1", "configuration": configuration},
             )
 
-    assert result.data["status"] == "failed"
-    assert message in result.data["error"]
     assert f"PATCH {INTEGRATION}" not in mock_router.paths()
 
 
@@ -643,13 +632,12 @@ async def test_updating_a_jira_configuration_is_refused(
     stub_integration(mock_router, JIRA_ATTRIBUTES)
 
     async with Client(mcp_root_server) as client:
-        result = await client.call_tool(
+        with pytest.raises(Exception, match="do not accept a configuration"):
+            await client.call_tool(
                 "prowler_update_integration",
                 {"integration_id": "i1", "configuration": {"domain": "other"}},
             )
 
-    assert result.data["status"] == "failed"
-    assert "do not accept a configuration" in result.data["error"]
     assert f"PATCH {INTEGRATION}" not in mock_router.paths()
 
 
@@ -660,12 +648,12 @@ async def test_attaching_a_jira_integration_to_a_provider_is_refused(
     stub_integration(mock_router, JIRA_ATTRIBUTES)
 
     async with Client(mcp_root_server) as client:
-        result = await client.call_tool(
+        with pytest.raises(Exception, match="tenant-wide"):
+            await client.call_tool(
                 "prowler_update_integration",
                 {"integration_id": "i1", "provider_ids": ["p1"]},
             )
 
-    assert "tenant-wide" in result.data["error"]
     assert f"PATCH {INTEGRATION}" not in mock_router.paths()
 
 
@@ -683,12 +671,12 @@ async def test_security_hub_must_keep_exactly_one_provider(
     stub_integration(mock_router, SECURITY_HUB_ATTRIBUTES, provider_ids=("p1",))
 
     async with Client(mcp_root_server) as client:
-        result = await client.call_tool(
+        with pytest.raises(Exception, match="exactly one AWS provider"):
+            await client.call_tool(
                 "prowler_update_integration",
                 {"integration_id": "i1", "provider_ids": provider_ids},
             )
 
-    assert "exactly one AWS provider" in result.data["error"]
     assert f"PATCH {INTEGRATION}" not in mock_router.paths()
 
 
@@ -708,12 +696,12 @@ async def test_partial_jira_credentials_are_refused_to_protect_the_stored_ones(
     stub_integration(mock_router, JIRA_ATTRIBUTES)
 
     async with Client(mcp_root_server) as client:
-        result = await client.call_tool(
+        with pytest.raises(Exception, match="replaced as a whole"):
+            await client.call_tool(
                 "prowler_update_integration",
                 {"integration_id": "i1", "credentials": credentials},
             )
 
-    assert "replaced as a whole" in result.data["error"]
     assert f"PATCH {INTEGRATION}" not in mock_router.paths()
 
 
@@ -751,13 +739,14 @@ async def test_replacing_jira_credentials_normalizes_the_domain(
 # ------------------------------------------------- delete and connection tools
 
 
-async def test_deleting_an_integration_reports_the_outcome_either_way(
+async def test_deleting_an_integration_confirms_it_happened(
     mcp_root_server, mock_api_client, mock_router
 ):
-    """Deletion is irreversible, so both outcomes are stated explicitly.
+    """Deletion is irreversible, so a success says so rather than staying silent.
 
-    A bare exception would leave the agent unsure whether the credentials are
+    It says so in the message and nowhere else: a `deleted: true` flag could only
-    gone, and a retry of a delete that actually succeeded reads as a new failure.
+    ever be true, because an integration that was not deleted leaves the tool as
+    an error.
     """
     mock_router.add("DELETE", INTEGRATION, status=204)
 
@@ -766,25 +755,24 @@ async def test_deleting_an_integration_reports_the_outcome_either_way(
             "prowler_delete_integration", {"integration_id": "i1"}
         )
 
-    assert result.data["deleted"] is True
+    assert "i1 deleted successfully" in result.data["message"]
+    assert "deleted" not in result.data
 
 
-async def test_a_failed_deletion_says_it_did_not_happen(
+async def test_a_refused_deletion_fails_and_says_the_role_is_the_problem(
     mcp_root_server, mock_api_client, mock_router
 ):
-    """`deleted: false` is the part the agent must not have to infer."""
+    """A 403 is the same answer for every tool, so `lib.errors` writes it."""
     mock_router.add(
         "DELETE", INTEGRATION, status=403, json=jsonapi_error(403, "Permission denied.")
     )
 
     async with Client(mcp_root_server) as client:
-        result = await client.call_tool(
+        with pytest.raises(Exception, match="prowler_get_current_user"):
+            await client.call_tool(
                 "prowler_delete_integration", {"integration_id": "i1"}
             )
 
-    assert result.data["deleted"] is False
-    assert "Permission denied." in result.data["message"]
-
 
 async def test_checking_a_connection_surfaces_why_it_failed(
     mcp_root_server, mock_api_client, mock_router
@@ -1026,6 +1014,32 @@ async def test_an_accepted_dispatch_with_no_task_id_is_not_safe_to_retry(
     assert "task_id" not in result.data
 
 
+async def test_a_dispatch_with_no_usable_credential_is_raised_not_called_unknown(
+    mcp_root_server, mock_api_client, mock_router, monkeypatch
+):
+    """Authentication runs before the request, so nothing was ever queued.
+
+    Reported as `unknown` it reads as "work items may exist in Jira, go and
+    look" -- for a call that never reached Prowler. It is not a dispatch outcome
+    at all: the credential has to be fixed, and no retry of this call does that.
+    """
+    monkeypatch.setattr(mock_api_client.auth_manager, "mode", "http")
+
+    async with Client(mcp_root_server) as client:
+        with pytest.raises(Exception, match="no usable credential"):
+            await client.call_tool(
+                "prowler_send_findings_to_jira",
+                {
+                    "integration_id": "i1",
+                    "project_key": "PROJ",
+                    "issue_type": "Task",
+                    "finding_ids": ["f1"],
+                },
+            )
+
+    assert mock_router.paths() == []
+
+
 async def test_a_dispatch_task_that_died_halfway_is_never_safe_to_retry(
     mcp_root_server, mock_api_client, mock_router
 ):

mcp_server/tests/prowler_app/tools/test_muting.py

@@ -0,0 +1,64 @@
+"""Tests for the muting tools.
+
+Muting is permanent and deleting a rule does not undo it, so the one thing
+these assertions protect is that an agent is never told a deletion failed when
+it did not: the old answer keyed off the *shape* of the API's reply rather than
+off anything having gone wrong, and said nothing a caller could act on.
+"""
+
+import pytest
+from fastmcp import Client
+
+from tests.helpers.jsonapi import jsonapi_document, jsonapi_error, jsonapi_resource
+
+MUTE_RULE = "/api/v1/mute-rules/m1"
+
+
+async def test_a_deleted_rule_is_reported_deleted_whatever_the_body(
+    mcp_root_server, mock_api_client, mock_router
+):
+    """Prowler answers 204 with no body, but 200 with one is just as much a yes.
+
+    The old check read ``success`` out of the parsed body, which only exists for
+    the empty-body case, so a deletion that worked could be reported as
+    "Failed to delete mute rule".
+    """
+    mock_router.add(
+        "DELETE",
+        MUTE_RULE,
+        json=jsonapi_document(jsonapi_resource("mute-rules", "m1", {})),
+    )
+
+    async with Client(mcp_root_server) as client:
+        result = await client.call_tool("prowler_delete_mute_rule", {"rule_id": "m1"})
+
+    assert "m1 deleted successfully" in result.data["message"]
+    assert "stay muted" in result.data["message"]
+    # A flag with one reachable value is not a fact, it is an invitation to
+    # branch on a shape that does not exist.
+    assert "success" not in result.data
+
+
+async def test_an_empty_body_deletion_is_reported_the_same_way(
+    mcp_root_server, mock_api_client, mock_router
+):
+    """The 204 path, which is what Prowler actually sends today."""
+    mock_router.add("DELETE", MUTE_RULE, status=204)
+
+    async with Client(mcp_root_server) as client:
+        result = await client.call_tool("prowler_delete_mute_rule", {"rule_id": "m1"})
+
+    assert "m1 deleted successfully" in result.data["message"]
+
+
+async def test_a_refused_deletion_is_an_error(
+    mcp_root_server, mock_api_client, mock_router
+):
+    """A rule that was not deleted has to leave the tool as an error."""
+    mock_router.add(
+        "DELETE", MUTE_RULE, status=404, json=jsonapi_error(404, "Not found.")
+    )
+
+    async with Client(mcp_root_server) as client:
+        with pytest.raises(Exception, match="Not found"):
+            await client.call_tool("prowler_delete_mute_rule", {"rule_id": "m1"})

mcp_server/tests/prowler_app/tools/test_providers.py

@@ -0,0 +1,316 @@
+"""Tests for the provider tools.
+
+Two behaviours drive most of the assertions here, and both are about telling a
+failure apart from an outcome that is merely not final yet:
+
+* Deleting a provider removes it together with its scans, findings and
+  resources, in a background task that routinely outlives the polling window.
+  A deletion still running is not a failure, and reporting it as one invites a
+  retry of a destructive call that is already in flight.
+* ``connect_provider`` runs a connection check, and a check that could not be
+  run says nothing about the provider's credentials. Reporting it as ``failed``
+  blames an AWS role for an expired Prowler credential and sends the user off to
+  fix something that works.
+
+Tools are driven through an in-memory MCP client so FastMCP resolves the
+pydantic ``Field`` defaults and a raised failure arrives the way a client sees
+it.
+"""
+
+import pytest
+from fastmcp import Client
+
+from tests.helpers.http import MockRouter
+from tests.helpers.jsonapi import (
+    jsonapi_collection,
+    jsonapi_document,
+    jsonapi_error,
+    jsonapi_resource,
+    task_document,
+)
+
+PROVIDERS = "/api/v1/providers"
+PROVIDER = f"{PROVIDERS}/p1"
+CONNECTION = f"{PROVIDER}/connection"
+SECRETS = f"{PROVIDERS}/secrets"
+TASK = "/api/v1/tasks/t1"
+
+PROVIDER_ATTRIBUTES = {
+    "uid": "123456789012",
+    "provider": "aws",
+    "alias": "production",
+    "connection": {"connected": True},
+}
+
+
+@pytest.fixture
+def mock_fast_polling(monkeypatch, api_client):
+    """Run the real polling loop, with a timeout a test can afford to wait out.
+
+    The timeout path is the one worth testing here -- it is what used to be
+    reported as a failed deletion -- so the loop, its exception and the fallback
+    that reads the task afterwards all stay real. Only the 60 seconds go.
+    """
+    original = type(api_client).poll_task_until_complete
+
+    async def _fast(self, task_id, **_overridden):
+        return await original(self, task_id, timeout=0.3, poll_interval=0.05)
+
+    monkeypatch.setattr(type(api_client), "poll_task_until_complete", _fast)
+
+
+@pytest.fixture
+def mock_polling_timeout(monkeypatch, api_client):
+    """Make the polling window run out on the first call, without the wait.
+
+    The clock is what ends the polling loop here, so a test about *what happens
+    afterwards* has no reason to spend it. The read the fallback then makes is
+    the real one.
+    """
+
+    async def _timeout(self, task_id, **_overridden):
+        raise TimeoutError(f"Task {task_id} polling timed out after 60 seconds.")
+
+    monkeypatch.setattr(type(api_client), "poll_task_until_complete", _timeout)
+
+
+def stub_deletion_start(mock_router: MockRouter) -> MockRouter:
+    """Serve the DELETE as Prowler does: a task to poll, not a finished deletion."""
+    return mock_router.add(
+        "DELETE", PROVIDER, json=jsonapi_document(jsonapi_resource("tasks", "t1", {}))
+    )
+
+
+# --------------------------------------------------------------- deletion
+
+
+async def test_a_refused_deletion_is_an_error_not_a_result(
+    mcp_root_server, mock_api_client, mock_router
+):
+    """Nothing started, so the classifier owns the message.
+
+    Returned as ``{"deleted": false}`` it arrives with ``isError: false`` and a
+    model has no reason to treat it as anything but a completed call.
+    """
+    mock_router.add(
+        "DELETE", PROVIDER, status=403, json=jsonapi_error(403, "Not allowed.")
+    )
+
+    async with Client(mcp_root_server) as client:
+        with pytest.raises(Exception, match="not allowed to do this"):
+            await client.call_tool("prowler_delete_provider", {"provider_id": "p1"})
+
+
+async def test_a_deletion_with_no_task_back_warns_before_a_blind_retry(
+    mcp_root_server, mock_api_client, mock_router
+):
+    """Prowler accepted it, so the deletion is probably running.
+
+    Without the task ID there is nothing to watch it with, which makes "check
+    whether it is gone" the only safe instruction.
+    """
+    mock_router.add("DELETE", PROVIDER, json={"data": {}})
+
+    async with Client(mcp_root_server) as client:
+        with pytest.raises(Exception, match="did not return the ID"):
+            await client.call_tool("prowler_delete_provider", {"provider_id": "p1"})
+
+    assert mock_router.paths() == [f"DELETE {PROVIDER}"]
+
+
+async def test_a_finished_deletion_reports_it_plainly(
+    mcp_root_server, mock_api_client, mock_router
+):
+    """The success path is unchanged: the provider is gone."""
+    stub_deletion_start(mock_router)
+    mock_router.add("GET", TASK, json=task_document("t1", "completed"))
+
+    async with Client(mcp_root_server) as client:
+        result = await client.call_tool(
+            "prowler_delete_provider", {"provider_id": "p1"}
+        )
+
+    assert result.data["status"] == "deleted"
+
+
+async def test_a_deletion_still_running_is_not_reported_as_a_failure(
+    mcp_root_server, mock_api_client, mock_router, mock_fast_polling
+):
+    """Outliving the polling window is normal for a provider with many findings.
+
+    The task ID goes back so the deletion can be followed, and the message says
+    not to send it again -- which is the whole point of not calling this failed.
+    """
+    stub_deletion_start(mock_router)
+    mock_router.add("GET", TASK, json=task_document("t1", "executing"))
+
+    async with Client(mcp_root_server) as client:
+        result = await client.call_tool(
+            "prowler_delete_provider", {"provider_id": "p1"}
+        )
+
+    assert result.data["status"] == "in_progress"
+    assert result.data["task_id"] == "t1"
+    assert "Do not send the deletion again" in result.data["message"]
+
+
+async def test_a_deletion_that_finished_just_after_the_wait_is_reported_as_deleted(
+    mcp_root_server, mock_api_client, mock_router, mock_polling_timeout
+):
+    """Polling gives up on the clock, not on the task.
+
+    A deletion that completed a moment after the last poll is a finished
+    deletion, and the read the fallback makes is what says so. Reporting it as
+    still running would send the caller off to watch a provider that is gone.
+    """
+    stub_deletion_start(mock_router)
+    mock_router.add("GET", TASK, json=task_document("t1", "completed"))
+
+    async with Client(mcp_root_server) as client:
+        result = await client.call_tool(
+            "prowler_delete_provider", {"provider_id": "p1"}
+        )
+
+    assert result.data["status"] == "deleted"
+    assert "task_id" not in result.data
+
+
+async def test_a_deletion_task_that_stopped_is_an_error_naming_what_is_left(
+    mcp_root_server, mock_api_client, mock_router
+):
+    """Here the provider really is still there, so this one is a failure.
+
+    A provider is removed together with everything attached to it, so a task
+    that stopped halfway can leave part of that gone -- which is why the message
+    sends the caller to look rather than asserting the state.
+    """
+    stub_deletion_start(mock_router)
+    mock_router.add("GET", TASK, json=task_document("t1", "failed", error="boom"))
+
+    async with Client(mcp_root_server) as client:
+        with pytest.raises(Exception, match="ended as 'failed'"):
+            await client.call_tool("prowler_delete_provider", {"provider_id": "p1"})
+
+
+async def test_a_deletion_task_failure_does_not_relay_the_upstream_text(
+    mcp_root_server, mock_api_client, mock_router
+):
+    """The task's own error is a celery traceback; it stays in the log."""
+    stub_deletion_start(mock_router)
+    mock_router.add(
+        "GET",
+        TASK,
+        json=task_document("t1", "failed", error="Traceback: secret-internal-detail"),
+    )
+
+    async with Client(mcp_root_server) as client:
+        with pytest.raises(Exception) as raised:
+            await client.call_tool("prowler_delete_provider", {"provider_id": "p1"})
+
+    assert "secret-internal-detail" not in str(raised.value)
+
+
+async def test_a_deletion_whose_progress_cannot_be_read_still_says_do_not_retry(
+    mcp_root_server, mock_api_client, mock_router, mock_fast_polling
+):
+    """The outcome is unknown, which for a destructive call means: do not repeat.
+
+    Reading the task is what tells "still running" from "stopped", so when that
+    read fails too the message drops the claim rather than guessing at one.
+    """
+    stub_deletion_start(mock_router)
+    mock_router.add("GET", TASK, status=503, json=jsonapi_error(503, "Unavailable."))
+
+    async with Client(mcp_root_server) as client:
+        result = await client.call_tool(
+            "prowler_delete_provider", {"provider_id": "p1"}
+        )
+
+    assert result.data["status"] == "in_progress"
+    assert "could not be read" in result.data["message"]
+    assert "Do not send the deletion again" in result.data["message"]
+    # The failure that got us here is the classifier's to phrase, so its raw
+    # text stays in the log rather than riding along in the message.
+    assert "API request failed" not in result.data["message"]
+
+
+# ------------------------------------------------------- connection check
+
+
+async def test_a_connection_check_that_cannot_run_is_not_reported_as_failed(
+    mcp_root_server, mock_api_client, mock_router
+):
+    """`not_tested` says nothing about the credentials, and that is the point.
+
+    A 401 here is this server's own credential, not the provider's. Calling it
+    `failed` tells the user their AWS role is broken when it is fine.
+    """
+    # Registered in order and consumed in order: the lookup before the create
+    # finds nothing, the one after it finds the provider that was just made.
+    mock_router.add("GET", PROVIDERS, json=jsonapi_collection([]))
+    mock_router.add(
+        "GET",
+        PROVIDERS,
+        json=jsonapi_collection(
+            [jsonapi_resource("providers", "p1", PROVIDER_ATTRIBUTES)]
+        ),
+    )
+    mock_router.add(
+        "POST",
+        PROVIDERS,
+        json=jsonapi_document(jsonapi_resource("providers", "p1", PROVIDER_ATTRIBUTES)),
+    )
+    mock_router.add(
+        "POST", CONNECTION, status=401, json=jsonapi_error(401, "Token expired.")
+    )
+    mock_router.add(
+        "GET",
+        PROVIDER,
+        json=jsonapi_document(jsonapi_resource("providers", "p1", PROVIDER_ATTRIBUTES)),
+    )
+
+    async with Client(mcp_root_server) as client:
+        result = await client.call_tool(
+            "prowler_connect_provider",
+            {"provider_uid": "123456789012", "provider_type": "aws"},
+        )
+
+    assert result.data["connected"] == "not_tested"
+    assert "never tested" in result.data["error"]
+
+
+async def test_a_secret_lookup_failure_does_not_pass_as_having_no_secret(
+    mcp_root_server, mock_api_client, mock_router
+):
+    """Returning None here would send the write down the create branch.
+
+    A provider holds at most one secret, so creating a second one is refused and
+    the caller would be told its credentials were rejected when all that
+    actually failed was this read.
+    """
+    mock_router.add(
+        "GET",
+        PROVIDERS,
+        json=jsonapi_collection(
+            [jsonapi_resource("providers", "p1", PROVIDER_ATTRIBUTES)]
+        ),
+    )
+    mock_router.add(
+        "GET", SECRETS, status=429, json=jsonapi_error(429, "Too many requests.")
+    )
+
+    async with Client(mcp_root_server) as client:
+        with pytest.raises(Exception, match="rate limiting"):
+            await client.call_tool(
+                "prowler_connect_provider",
+                {
+                    "provider_uid": "123456789012",
+                    "provider_type": "aws",
+                    "credentials": {
+                        "aws_access_key_id": "AKIA",
+                        "aws_secret_access_key": "s",
+                    },
+                },
+            )
+
+    assert f"POST {SECRETS}" not in mock_router.paths()

mcp_server/tests/prowler_app/tools/test_roles.py

@@ -0,0 +1,147 @@
+"""Tests for the role (RBAC) tools.
+
+``prowler_set_user_role`` replaces the single role a user holds, and the API
+silently drops a role ID that does not exist in the tenant -- which would leave
+the user with no role at all. So the tool reads the role first, and what that
+read says has to be told apart carefully: only a not-found is about the role ID
+the caller passed. A permission error, a rate limit or a server error is about
+the request, and reporting either as "find a valid role ID" sends the user to
+fix an ID that was fine.
+
+Tools are driven through an in-memory MCP client so FastMCP resolves the
+pydantic ``Field`` defaults and a raised failure arrives the way a client sees
+it.
+"""
+
+import pytest
+from fastmcp import Client
+
+from tests.helpers.http import MockRouter
+from tests.helpers.jsonapi import (
+    jsonapi_document,
+    jsonapi_error,
+    jsonapi_resource,
+)
+
+USER = "/api/v1/users/u1"
+USER_ROLES = f"{USER}/relationships/roles"
+ROLE = "/api/v1/roles/r2"
+
+ROLE_ATTRIBUTES = {"name": "admin", "manage_account": True}
+
+
+def stub_user_holding(mock_router: MockRouter, role_id: str) -> MockRouter:
+    """Serve ``GET /users/u1?include=roles`` with the user holding one role."""
+    return mock_router.add(
+        "GET",
+        USER,
+        json=jsonapi_document(
+            jsonapi_resource("users", "u1", {"name": "Ada"}),
+            included=[jsonapi_resource("roles", role_id, ROLE_ATTRIBUTES)],
+        ),
+    )
+
+
+async def test_setting_a_role_the_user_already_holds_changes_nothing(
+    mcp_root_server, mock_api_client, mock_router
+):
+    """Idempotent by design: no PATCH goes out, so nothing can be replaced."""
+    stub_user_holding(mock_router, "r2")
+
+    async with Client(mcp_root_server) as client:
+        result = await client.call_tool(
+            "prowler_set_user_role", {"user_id": "u1", "role_id": "r2"}
+        )
+
+    assert result.data["changed"] is False
+    assert mock_router.paths() == [f"GET {USER}"]
+
+
+async def test_a_role_that_does_not_exist_is_named_as_the_bad_argument(
+    mcp_root_server, mock_api_client, mock_router
+):
+    """404 is the one answer that really is about the role ID.
+
+    The PATCH would accept the ID and drop it, leaving the user with no role, so
+    the read has to stop the call -- and say which ID to replace.
+    """
+    stub_user_holding(mock_router, "r1")
+    mock_router.add("GET", ROLE, status=404, json=jsonapi_error(404, "Not found."))
+
+    async with Client(mcp_root_server) as client:
+        with pytest.raises(Exception, match="does not exist in this tenant") as raised:
+            await client.call_tool(
+                "prowler_set_user_role", {"user_id": "u1", "role_id": "r2"}
+            )
+
+    assert "left unchanged" in str(raised.value)
+    assert f"PATCH {USER_ROLES}" not in mock_router.paths()
+
+
+@pytest.mark.parametrize(
+    ("status", "detail", "expected"),
+    [
+        (403, "Not allowed.", "not allowed to do this"),
+        (429, "Slow down.", "rate limiting"),
+        (500, "Boom.", "failed on Prowler's side"),
+    ],
+    ids=["forbidden", "rate-limited", "server-error"],
+)
+async def test_a_role_read_that_failed_for_another_reason_is_not_a_bad_role_id(
+    mcp_root_server, mock_api_client, mock_router, status, detail, expected
+):
+    """These say nothing about the ID, so the classifier owns the message.
+
+    Told "use prowler_list_roles to find a valid role ID", an agent goes looking
+    for a role that was never the problem -- and finds the same wall.
+    """
+    stub_user_holding(mock_router, "r1")
+    mock_router.add("GET", ROLE, status=status, json=jsonapi_error(status, detail))
+
+    async with Client(mcp_root_server) as client:
+        with pytest.raises(Exception, match=expected) as raised:
+            await client.call_tool(
+                "prowler_set_user_role", {"user_id": "u1", "role_id": "r2"}
+            )
+
+    assert "valid role ID" not in str(raised.value)
+    assert f"PATCH {USER_ROLES}" not in mock_router.paths()
+
+
+async def test_a_set_role_reports_the_role_the_user_holds_afterwards(
+    mcp_root_server, mock_api_client, mock_router
+):
+    """The authoritative state is read back rather than assumed from the PATCH."""
+    mock_router.add(
+        "GET",
+        USER,
+        json=jsonapi_document(
+            jsonapi_resource("users", "u1", {"name": "Ada"}),
+            included=[jsonapi_resource("roles", "r1", ROLE_ATTRIBUTES)],
+        ),
+    )
+    mock_router.add(
+        "GET",
+        USER,
+        json=jsonapi_document(
+            jsonapi_resource("users", "u1", {"name": "Ada"}),
+            included=[jsonapi_resource("roles", "r2", ROLE_ATTRIBUTES)],
+        ),
+    )
+    mock_router.add(
+        "GET",
+        ROLE,
+        json=jsonapi_document(jsonapi_resource("roles", "r2", ROLE_ATTRIBUTES)),
+    )
+    mock_router.add("PATCH", USER_ROLES, status=204)
+
+    async with Client(mcp_root_server) as client:
+        result = await client.call_tool(
+            "prowler_set_user_role", {"user_id": "u1", "role_id": "r2"}
+        )
+
+    assert result.data["changed"] is True
+    assert [role["id"] for role in result.data["roles"]] == ["r2"]
+    assert mock_router.json_body("PATCH", USER_ROLES) == {
+        "data": [{"type": "roles", "id": "r2"}]
+    }

mcp_server/tests/prowler_app/tools/test_scans.py

@@ -0,0 +1,201 @@
+"""Tests for the scans tools.
+
+Both write tools here used to answer a failure with a result object that the
+protocol, the client and the model all read as a success, and both are calls
+whose outcome an agent acts on:
+
+* ``prowler_trigger_scan`` starts work. Anything that reads as "nothing
+  happened" invites a second scan of the same provider.
+* ``prowler_schedule_daily_scan`` was deciding whether the schedule existed by
+  reading the state of a different thing entirely -- the first scan Prowler
+  starts alongside it -- so a schedule that had just been created could be
+  reported as a failure. Retrying that can only hit the 409 the API raises for a
+  provider that already has one.
+
+Tools are driven through an in-memory MCP client so FastMCP resolves the
+pydantic ``Field`` defaults and a raised failure arrives the way a client sees
+it.
+"""
+
+import pytest
+from fastmcp import Client
+
+from tests.helpers.http import MockRouter
+from tests.helpers.jsonapi import (
+    jsonapi_document,
+    jsonapi_error,
+    jsonapi_resource,
+)
+
+SCANS = "/api/v1/scans"
+SCAN = f"{SCANS}/s1"
+DAILY = "/api/v1/schedules/daily"
+
+SCAN_ATTRIBUTES = {
+    "name": "Nightly",
+    "trigger": "manual",
+    "state": "executing",
+    "progress": 40,
+}
+
+
+def stub_scan_creation(mock_router: MockRouter, scan_id: str = "s1") -> MockRouter:
+    """Serve the creation as Prowler does: a task carrying the new scan's ID."""
+    return mock_router.add(
+        "POST",
+        SCANS,
+        json=jsonapi_document(
+            jsonapi_resource("tasks", "t1", {"task_args": {"scan_id": scan_id}})
+        ),
+    )
+
+
+# ------------------------------------------------------------- trigger_scan
+
+
+async def test_a_refused_scan_is_an_error_not_a_failed_looking_result(
+    mcp_root_server, mock_api_client, mock_router
+):
+    """A rejection has to leave the tool as an error.
+
+    Returned as a result it arrives with ``isError: false``, and a model reading
+    a successful tool call has no reason to doubt that a scan is now running.
+    """
+    mock_router.add(
+        "POST", SCANS, status=403, json=jsonapi_error(403, "Insufficient permissions.")
+    )
+
+    async with Client(mcp_root_server) as client:
+        with pytest.raises(Exception, match="not allowed to do this"):
+            await client.call_tool("prowler_trigger_scan", {"provider_id": "p1"})
+
+
+async def test_a_scan_with_no_id_back_warns_before_a_blind_retry(
+    mcp_root_server, mock_api_client, mock_router
+):
+    """Prowler accepted it, so a second call could start a duplicate scan.
+
+    The message names the provider because that is what makes the suggested
+    check actionable without another lookup.
+    """
+    mock_router.add("POST", SCANS, json={"data": {"attributes": {"task_args": {}}}})
+
+    async with Client(mcp_root_server) as client:
+        with pytest.raises(Exception, match="did not return its ID"):
+            await client.call_tool("prowler_trigger_scan", {"provider_id": "p1"})
+
+    assert mock_router.paths() == [f"POST {SCANS}"]
+
+
+async def test_a_scan_whose_read_back_fails_still_hands_over_the_id(
+    mcp_root_server, mock_api_client, mock_router
+):
+    """The scan is running; only reading it back went wrong.
+
+    Reporting the read failure alone would read as "the scan was not created"
+    and invite a duplicate, so the error carries the ID to monitor instead.
+    """
+    stub_scan_creation(mock_router)
+    mock_router.add("GET", SCAN, status=400, json=jsonapi_error(400, "Server error."))
+
+    async with Client(mcp_root_server) as client:
+        with pytest.raises(Exception, match="Scan s1 was created") as raised:
+            await client.call_tool("prowler_trigger_scan", {"provider_id": "p1"})
+
+    # Naming the scan is the whole reason this message exists, so it is written
+    # here rather than assembled from the failure. Splicing the failure text in
+    # would put whatever it happens to say in front of the model unclassified,
+    # which is the one thing the shared classifier exists to decide.
+    assert "API request failed" not in str(raised.value)
+
+
+async def test_a_created_scan_comes_back_with_its_details(
+    mcp_root_server, mock_api_client, mock_router
+):
+    """The success path still reports the scan, which is what gets monitored."""
+    stub_scan_creation(mock_router)
+    mock_router.add(
+        "GET",
+        SCAN,
+        json=jsonapi_document(jsonapi_resource("scans", "s1", SCAN_ATTRIBUTES)),
+    )
+
+    async with Client(mcp_root_server) as client:
+        result = await client.call_tool("prowler_trigger_scan", {"provider_id": "p1"})
+
+    assert result.data["scan"]["id"] == "s1"
+    # No status flag: a scan that was not created is raised, so "success" could
+    # only ever be the one value and says nothing a reader can act on.
+    assert "status" not in result.data
+
+
+# ------------------------------------------------------ schedule_daily_scan
+
+
+@pytest.mark.parametrize("first_run_state", ["available", "scheduled", "executing"])
+async def test_a_schedule_is_reported_created_whatever_the_first_run_does(
+    mcp_root_server, mock_api_client, mock_router, first_run_state
+):
+    """The task in the answer is the first scan run, not the schedule.
+
+    Prowler commits the recurring schedule inside the request that serves this
+    call, so an answer at all means it exists. Reading that task's state as the
+    outcome of the scheduling reported a schedule that had just been created as
+    a failure.
+    """
+    mock_router.add(
+        "POST",
+        DAILY,
+        json=jsonapi_document(
+            jsonapi_resource("tasks", "t1", {"state": first_run_state})
+        ),
+    )
+
+    async with Client(mcp_root_server) as client:
+        result = await client.call_tool(
+            "prowler_schedule_daily_scan", {"provider_id": "p1"}
+        )
+
+    assert result.data["first_run_state"] == first_run_state
+    assert "every 24 hours" in result.data["message"]
+    assert "scheduled" not in result.data
+
+
+async def test_a_failed_first_run_leaves_the_schedule_standing(
+    mcp_root_server, mock_api_client, mock_router
+):
+    """Worth saying, but it is not the schedule that failed.
+
+    The gap it leaves is real -- no findings until tomorrow -- so the message
+    points at the manual scan that fills it rather than at the scheduling.
+    """
+    mock_router.add(
+        "POST",
+        DAILY,
+        json=jsonapi_document(jsonapi_resource("tasks", "t1", {"state": "failed"})),
+    )
+
+    async with Client(mcp_root_server) as client:
+        result = await client.call_tool(
+            "prowler_schedule_daily_scan", {"provider_id": "p1"}
+        )
+
+    assert result.data["first_run_state"] == "failed"
+    assert "schedule is unaffected" in result.data["message"]
+    assert "prowler_trigger_scan" in result.data["message"]
+
+
+async def test_an_existing_schedule_is_relayed_as_the_api_explains_it(
+    mcp_root_server, mock_api_client, mock_router
+):
+    """The 409 already says the useful thing, so the classifier relays it."""
+    mock_router.add(
+        "POST",
+        DAILY,
+        status=409,
+        json=jsonapi_error(409, "There is already a scheduled scan for this provider."),
+    )
+
+    async with Client(mcp_root_server) as client:
+        with pytest.raises(Exception, match="already a scheduled scan"):
+            await client.call_tool("prowler_schedule_daily_scan", {"provider_id": "p1"})

mcp_server/tests/prowler_app/utils/test_auth.py

@@ -6,8 +6,12 @@ Reference for later branches: ``ProwlerAppAuth`` resolves its ``mode`` and
 and ``base_url=`` explicitly, as these tests do.
 """
 
+import base64
+import json
+
 import pytest
 
+from prowler_mcp_server.lib.errors import CredentialError
 from prowler_mcp_server.prowler_app.utils.auth import ProwlerAppAuth
 from tests.helpers.tokens import FAKE_API_KEY, MALFORMED_API_KEY, fake_jwt
 
@@ -42,13 +46,92 @@ async def test_http_mode_accepts_a_bearer_api_key(http_request_headers):
     assert await auth.get_valid_token() == FAKE_API_KEY
 
 
+def _jwt_with_payload(payload: object) -> str:
+    """Mint an unsigned JWT carrying an arbitrary payload.
+
+    ``fake_jwt`` always writes a well-formed object, so the malformed payloads
+    below are built here instead.
+    """
+    encoded = (
+        base64.urlsafe_b64encode(json.dumps(payload).encode()).decode().rstrip("=")
+    )
+    return f"header.{encoded}.fake-signature-not-verified"
+
+
+async def test_http_mode_accepts_a_lowercase_bearer_scheme(http_request_headers):
+    """Authentication scheme names are case-insensitive (RFC 7235)."""
+    http_request_headers(authorization=f"bearer {FAKE_API_KEY}")
+
+    auth = ProwlerAppAuth(mode="http")
+
+    assert await auth.get_valid_token() == FAKE_API_KEY
+
+
+async def test_http_mode_strips_only_the_scheme_prefix(http_request_headers):
+    """A token that repeats the scheme keeps it: only the prefix is removed."""
+    token = f"{FAKE_API_KEY}_Bearer_suffix"
+    http_request_headers(authorization=f"Bearer {token}")
+
+    auth = ProwlerAppAuth(mode="http")
+
+    assert await auth.get_valid_token() == token
+
+
+async def test_http_mode_rejects_an_authorization_header_without_a_token(
+    http_request_headers,
+):
+    """A bare scheme carries no credential to authenticate with."""
+    http_request_headers(authorization="Bearer   ")
+
+    auth = ProwlerAppAuth(mode="http")
+
+    with pytest.raises(CredentialError, match="'Bearer <token>' form"):
+        await auth.get_valid_token()
+
+
+async def test_http_mode_rejects_a_jwt_whose_payload_is_not_an_object(
+    http_request_headers,
+):
+    """A payload that decodes to a list has no claims, so it is a bad credential.
+
+    Without the type check it would reach `payload.get` and fail as an
+    unclassified `AttributeError`, which the client only sees masked.
+    """
+    http_request_headers(authorization=f"Bearer {_jwt_with_payload(['exp'])}")
+
+    auth = ProwlerAppAuth(mode="http")
+
+    with pytest.raises(CredentialError, match="not a readable JWT"):
+        await auth.get_valid_token()
+
+
+@pytest.mark.parametrize(
+    ("payload", "case"),
+    [
+        ({"sub": "user"}, "missing"),
+        ({"exp": "1700000000"}, "string"),
+        ({"exp": None}, "null"),
+    ],
+)
+async def test_http_mode_rejects_a_jwt_without_a_numeric_expiration(
+    http_request_headers, payload: dict, case: str
+):
+    """`exp` is a numeric date: comparing anything else raises a `TypeError`."""
+    http_request_headers(authorization=f"Bearer {_jwt_with_payload(payload)}")
+
+    auth = ProwlerAppAuth(mode="http")
+
+    with pytest.raises(CredentialError, match="no readable 'exp' expiration claim"):
+        await auth.get_valid_token()
+
+
 async def test_http_mode_rejects_an_expired_jwt(http_request_headers):
     """An expired JWT is refused locally instead of being forwarded to the API."""
     http_request_headers(authorization=f"Bearer {fake_jwt(expires_in=-60)}")
 
     auth = ProwlerAppAuth(mode="http")
 
-    with pytest.raises(ValueError, match="Token has expired"):
+    with pytest.raises(CredentialError, match="The token has expired"):
         await auth.get_valid_token()