Add suggestions by CodeRabbit review

Author: luis5tb

src/app/endpoints/query_v2.py

@@ -191,7 +191,7 @@ async def retrieve_response(  # pylint: disable=too-many-locals,too-many-branche
 
     # Prepare tools for responses API
     toolgroups = await prepare_tools_for_responses_api(
-        client, query_request, token, configuration
+        client, query_request, token, configuration, mcp_headers
     )
 
     # Prepare input for Responses API
@@ -455,6 +455,7 @@ async def prepare_tools_for_responses_api(
     query_request: QueryRequest,
     token: str,
     config: AppConfig,
+    mcp_headers: dict[str, dict[str, str]] | None = None,
 ) -> list[dict[str, Any]] | None:
     """
     Prepare tools for Responses API including RAG and MCP tools.
@@ -467,6 +468,7 @@ async def prepare_tools_for_responses_api(
         query_request: The user's query request
         token: Authentication token for MCP tools
         config: Configuration object containing MCP server settings
+        mcp_headers: Per-request headers for MCP servers
 
     Returns:
         list[dict[str, Any]] | None: List of tool configurations for the
@@ -489,6 +491,13 @@ async def prepare_tools_for_responses_api(
     # Add MCP server tools
     mcp_tools = get_mcp_tools(config.mcp_servers, token)
     if mcp_tools:
+        # Merge per-request headers if provided
+        if mcp_headers:
+            for tool in mcp_tools:
+                server_url = tool.get("server_url")
+                if server_url and server_url in mcp_headers:
+                    tool_headers = tool.setdefault("headers", {})
+                    tool_headers.update(mcp_headers[server_url])
         toolgroups.extend(mcp_tools)
         logger.debug(
             "Configured %d MCP tools: %s",

src/app/endpoints/streaming_query_v2.py

@@ -171,21 +171,13 @@ async def response_generator(
             event_type = getattr(chunk, "type", None)
             logger.debug("Processing chunk %d, type: %s", chunk_id, event_type)
 
-            # Emit start and persist on response.created
+            # Emit start on response.created
             if event_type == "response.created":
                 try:
                     conv_id = getattr(chunk, "response").id
                 except Exception:  # pylint: disable=broad-except
                     conv_id = ""
                 yield stream_start_event(conv_id)
-                if conv_id:
-                    persist_user_conversation_details(
-                        user_id=user_id,
-                        conversation_id=conv_id,
-                        model=model_id,
-                        provider_id=provider_id,
-                        topic_summary=None,
-                    )
                 continue
 
             # Text streaming
@@ -448,11 +440,21 @@ async def retrieve_response(
 
     # Prepare tools for responses API
     toolgroups = await prepare_tools_for_responses_api(
-        client, query_request, token, configuration
+        client, query_request, token, configuration, mcp_headers
     )
 
+    # Prepare input for Responses API
+    # Convert attachments to text and concatenate with query
+    input_text = query_request.query
+    if query_request.attachments:
+        for attachment in query_request.attachments:
+            input_text += (
+                f"\n\n[Attachment: {attachment.attachment_type}]\n"
+                f"{attachment.content}"
+            )
+
     response = await client.responses.create(
-        input=query_request.query,
+        input=input_text,
         model=model_id,
         instructions=system_prompt,
         previous_response_id=query_request.conversation_id,

tests/unit/app/endpoints/test_streaming_query_v2.py

@@ -212,8 +212,8 @@ async def fake_stream():
     assert events[-1] == "END\n"
 
     # Verify conversation persistence was invoked with the created id
-    # Called twice: once on response.created, once at the end
-    assert persist_spy.call_count == 2
+    # Called once at the end (after topic summary logic)
+    assert persist_spy.call_count == 1
     persist_spy.assert_called_with(
         user_id="user123",
         conversation_id="conv-xyz",