fix(mypy): fix presidio, panw, perplexity, and mcp hook type issues

Task 6: Fix Presidio guardrail type issues (5 errors)
- Cast event_hook lists to List[GuardrailEventHooks]
- Cast response to dict for _process_anthropic_response_for_pii
- Remove bytes from async_post_call_streaming_iterator_hook return type

Task 7: Fix PANW Prisma AIRS type issues (3 errors)
- Annotate contents as List[Dict[str, Any]]
- Add type annotation and type: ignore for error_obj dict

Task 8: Fix Perplexity responses type issues (5 errors)
- Change _ensure_message_type return type to Union[str, ResponseInputParam]
- Add explicit List[Any] annotation for result

Task 9: Fix MCP semantic filter hook override (1 error)
- Add litellm_call_info parameter to match superclass signature

Co-authored-by: yuneng-jiang <yuneng-jiang@users.noreply.github.com>

litellm/llms/perplexity/responses/transformation.py

@@ -63,12 +63,12 @@ class PerplexityResponsesConfig(OpenAIResponsesAPIConfig):
 
     def _ensure_message_type(
         self, input: Union[str, ResponseInputParam]
-    ) -> Union[str, List[Dict[str, Any]]]:
+    ) -> Union[str, ResponseInputParam]:
         """Ensure list input items have type='message' (required by Perplexity)."""
         if isinstance(input, str):
             return input
         if isinstance(input, list):
-            result = []
+            result: List[Any] = []
             for item in input:
                 if isinstance(item, dict) and "type" not in item:
                     item = {**item, "type": "message"}

litellm/proxy/guardrails/guardrail_hooks/panw_prisma_airs/panw_prisma_airs.py

@@ -315,6 +315,7 @@ class PanwPrismaAirsHandler(CustomGuardrail):
             panw_metadata["litellm_trace_id"] = metadata["litellm_trace_id"]
 
         # Build contents: tool_event takes priority, else prompt/response text
+        contents: List[Dict[str, Any]]
         if tool_event is not None:
             contents = [{"tool_event": tool_event}]
         else:
@@ -1485,7 +1486,7 @@ class PanwPrismaAirsHandler(CustomGuardrail):
             detail = (
                 e.detail if isinstance(e.detail, dict) else {"message": str(e.detail)}
             )
-            error_obj = dict(detail.get("error", detail))
+            error_obj: Dict[str, Any] = dict(detail.get("error", detail))  # type: ignore[arg-type]
             error_obj["code"] = e.status_code
             yield f"data: {json.dumps({'error': error_obj})}\n\n"
         except Exception as e:

litellm/proxy/guardrails/guardrail_hooks/presidio.py

@@ -106,9 +106,9 @@ class _OPTIONAL_PresidioPIIMasking(CustomGuardrail):
         if (self.output_parse_pii or self.apply_to_output) and not logging_only:
             current_hook = self.event_hook
             if isinstance(current_hook, str) and current_hook != "post_call":
-                self.event_hook = [current_hook, "post_call"]
+                self.event_hook = cast(List[GuardrailEventHooks], [current_hook, "post_call"])
             elif isinstance(current_hook, list) and "post_call" not in current_hook:
-                self.event_hook = current_hook + ["post_call"]
+                self.event_hook = cast(List[GuardrailEventHooks], current_hook + ["post_call"])
         self.pii_entities_config: Dict[Union[PiiEntityType, str], PiiAction] = (
             pii_entities_config or {}
         )
@@ -908,7 +908,7 @@ class _OPTIONAL_PresidioPIIMasking(CustomGuardrail):
         if self.apply_to_output is True:
             if self._is_anthropic_message_response(response):
                 return await self._process_anthropic_response_for_pii(
-                    response=response, request_data=data, mode="mask"
+                    response=cast(dict, response), request_data=data, mode="mask"
                 )
             return await self._mask_output_response(
                 response=response, request_data=data
@@ -927,7 +927,7 @@ class _OPTIONAL_PresidioPIIMasking(CustomGuardrail):
             )
         elif self._is_anthropic_message_response(response):
             await self._process_anthropic_response_for_pii(
-                response=response, request_data=data, mode="unmask"
+                response=cast(dict, response), request_data=data, mode="unmask"
             )
         return response
 
@@ -1234,7 +1234,7 @@ class _OPTIONAL_PresidioPIIMasking(CustomGuardrail):
         user_api_key_dict: UserAPIKeyAuth,
         response: Any,
         request_data: dict,
-    ) -> AsyncGenerator[Union[ModelResponseStream, bytes], None]:
+    ) -> AsyncGenerator[ModelResponseStream, None]:
         """
         Process streaming response chunks to unmask PII tokens when needed.
         """

litellm/proxy/hooks/mcp_semantic_filter/hook.py

@@ -259,6 +259,7 @@ class SemanticToolFilterHook(CustomLogger):
         user_api_key_dict: "UserAPIKeyAuth",
         response: Any,
         request_headers: Optional[Dict[str, str]] = None,
+        litellm_call_info: Optional[Dict[str, Any]] = None,
     ) -> Optional[Dict[str, str]]:
         """Add semantic filter stats and tool names to response headers."""
         from litellm.constants import MAX_MCP_SEMANTIC_FILTER_TOOLS_HEADER_LENGTH