fix(mcp): preserve original response chain on follow-up

litellm/responses/main.py

@@ -356,7 +356,7 @@ async def aresponses_api_with_mcp(
                     follow_up_input=follow_up_input,
                     model=model,
                     all_tools=all_tools,
-                    response_id=response.id,
+                    previous_response_id=previous_response_id,
                     **follow_up_call_params,
                 )
 

litellm/responses/mcp/litellm_proxy_mcp_handler.py

@@ -1005,7 +1005,7 @@ class LiteLLM_Proxy_MCP_Handler:
         follow_up_input: List[Any],
         model: str,
         all_tools: Optional[List[Any]],
-        response_id: str,
+        previous_response_id: Optional[str],
         **call_params: Any,
     ) -> Union[ResponsesAPIResponse, BaseResponsesAPIStreamingIterator]:
         """Make follow-up response API call with tool results."""
@@ -1013,7 +1013,7 @@ class LiteLLM_Proxy_MCP_Handler:
             input=follow_up_input,
             model=model,
             tools=all_tools,  # Keep tools for potential future calls
-            previous_response_id=response_id,  # Link to previous response
+            previous_response_id=previous_response_id,
             **call_params,
         )
 

tests/test_litellm/responses/mcp/test_litellm_proxy_mcp_handler.py

@@ -103,9 +103,7 @@ def test_extract_tool_calls_from_chat_response_handles_tool_calls():
         object="chat.completion",
     )
 
-    tool_calls = LiteLLM_Proxy_MCP_Handler._extract_tool_calls_from_chat_response(
-        response
-    )
+    tool_calls = LiteLLM_Proxy_MCP_Handler._extract_tool_calls_from_chat_response(response)
 
     assert len(tool_calls) == 1
     assert tool_calls[0]["function"]["name"] == "foo"
@@ -175,9 +173,7 @@ def test_transform_mcp_tools_to_openai_uses_chat_format(monkeypatch):
         fake_transform_responses,
     )
 
-    chat_tools = LiteLLM_Proxy_MCP_Handler._transform_mcp_tools_to_openai(
-        ["tool"], target_format="chat"
-    )
+    chat_tools = LiteLLM_Proxy_MCP_Handler._transform_mcp_tools_to_openai(["tool"], target_format="chat")
     resp_tools = LiteLLM_Proxy_MCP_Handler._transform_mcp_tools_to_openai(["tool"])
 
     assert chat_tools == [{"chat": True}]
@@ -216,6 +212,35 @@ def test_create_follow_up_input_handles_response_function_tool_call():
     ]
 
 
+@pytest.mark.asyncio
+@pytest.mark.parametrize("previous_response_id", [None, "resp_previous"])
+async def test_make_follow_up_call_uses_original_previous_response_id(
+    monkeypatch: pytest.MonkeyPatch,
+    previous_response_id: str | None,
+):
+    expected_response = MagicMock()
+    aresponses_mock = AsyncMock(return_value=expected_response)
+    handler_module = importlib.import_module("litellm.responses.mcp.litellm_proxy_mcp_handler")
+    monkeypatch.setattr(handler_module, "aresponses", aresponses_mock)
+
+    response = await LiteLLM_Proxy_MCP_Handler._make_follow_up_call(
+        follow_up_input=[
+            {
+                "type": "function_call_output",
+                "call_id": "call-1",
+                "output": "done",
+            }
+        ],
+        model="test-model",
+        all_tools=[],
+        previous_response_id=previous_response_id,
+    )
+
+    assert response is expected_response
+    assert aresponses_mock.await_args is not None
+    assert aresponses_mock.await_args.kwargs["previous_response_id"] == previous_response_id
+
+
 @pytest.mark.asyncio
 async def test_execute_tool_calls_strips_server_prefix(monkeypatch):
     call_tool_mock = _setup_mcp_call_environment(monkeypatch)
@@ -297,9 +322,7 @@ async def test_execute_tool_calls_strips_prefix_when_alias_differs_from_server_n
     )
     from litellm.proxy._experimental.mcp_server import mcp_server_manager as _msm
 
-    _msm.global_mcp_server_manager._get_mcp_server_from_tool_name = MagicMock(
-        return_value=fake_server
-    )
+    _msm.global_mcp_server_manager._get_mcp_server_from_tool_name = MagicMock(return_value=fake_server)
 
     tool_name = "my_deepwiki-read_wiki_structure"
     tool_calls = [
@@ -371,18 +394,14 @@ async def test_execute_tool_calls_logs_failure_via_post_call_failure_hook(monkey
     """
     post_call_failure_hook = _setup_proxy_logging(monkeypatch)
 
-    fake_manager = types.SimpleNamespace(
-        call_tool=AsyncMock(side_effect=HTTPException(status_code=500, detail="boom"))
-    )
+    fake_manager = types.SimpleNamespace(call_tool=AsyncMock(side_effect=HTTPException(status_code=500, detail="boom")))
     monkeypatch.setattr(
         "litellm.proxy._experimental.mcp_server.mcp_server_manager.global_mcp_server_manager",
         fake_manager,
     )
 
     tool_name = "deepwiki-read_wiki_structure"
-    tool_calls = [
-        {"id": "call-err", "function": {"name": tool_name, "arguments": "{}"}}
-    ]
+    tool_calls = [{"id": "call-err", "function": {"name": tool_name, "arguments": "{}"}}]
 
     user_auth = types.SimpleNamespace(api_key="test_key", user_id="test_user")
 
@@ -400,10 +419,7 @@ async def test_execute_tool_calls_logs_failure_via_post_call_failure_hook(monkey
 
     post_call_failure_hook.assert_awaited_once()
     assert post_call_failure_hook.await_args is not None
-    assert (
-        post_call_failure_hook.await_args.kwargs.get("route")
-        == "/responses/mcp/call_tool"
-    )
+    assert post_call_failure_hook.await_args.kwargs.get("route") == "/responses/mcp/call_tool"
 
 
 @pytest.mark.asyncio
@@ -426,9 +442,7 @@ async def test_execute_tool_calls_passes_litellm_call_id_and_trace_id_to_functio
     # NOTE: Don't patch via dotted string path here because `litellm.responses`
     # is a function attribute on the `litellm` package (shadowing the submodule),
     # which breaks monkeypatch's importpath resolution.
-    handler_module = importlib.import_module(
-        "litellm.responses.mcp.litellm_proxy_mcp_handler"
-    )
+    handler_module = importlib.import_module("litellm.responses.mcp.litellm_proxy_mcp_handler")
     monkeypatch.setattr(handler_module, "function_setup", fake_function_setup)
 
     tool_name = "deepwiki-read_wiki_structure"
@@ -475,9 +489,7 @@ async def test_get_mcp_tools_from_manager_enables_list_tools_logging(monkeypatch
     user_auth = types.SimpleNamespace(api_key="test_key", user_id="test_user")
     tools, _server_names = await LiteLLM_Proxy_MCP_Handler._get_mcp_tools_from_manager(
         user_api_key_auth=user_auth,
-        mcp_tools_with_litellm_proxy=[
-            {"type": "mcp", "server_url": "litellm_proxy/mcp/deepwiki"}
-        ],
+        mcp_tools_with_litellm_proxy=[{"type": "mcp", "server_url": "litellm_proxy/mcp/deepwiki"}],
     )
 
     assert tools == []
@@ -488,9 +500,7 @@ async def test_get_mcp_tools_from_manager_enables_list_tools_logging(monkeypatch
 
 
 def test_get_parent_request_tags_from_metadata():
-    tags = LiteLLM_Proxy_MCP_Handler._get_parent_request_tags(
-        {"metadata": {"tags": ["team-a", "prod"]}}
-    )
+    tags = LiteLLM_Proxy_MCP_Handler._get_parent_request_tags({"metadata": {"tags": ["team-a", "prod"]}})
     assert tags == ["team-a", "prod"]
 
 
@@ -526,9 +536,7 @@ async def test_get_mcp_tools_from_manager_forwards_request_tags(monkeypatch):
 
     await LiteLLM_Proxy_MCP_Handler._get_mcp_tools_from_manager(
         user_api_key_auth=types.SimpleNamespace(api_key="k", user_id="u"),
-        mcp_tools_with_litellm_proxy=[
-            {"type": "mcp", "server_url": "litellm_proxy/mcp/deepwiki"}
-        ],
+        mcp_tools_with_litellm_proxy=[{"type": "mcp", "server_url": "litellm_proxy/mcp/deepwiki"}],
         request_tags=["team-a"],
     )
 
@@ -545,9 +553,7 @@ async def test_execute_tool_calls_propagates_request_tags_to_function_setup(monk
         captured.update(kwargs)
         return None, None
 
-    handler_module = importlib.import_module(
-        "litellm.responses.mcp.litellm_proxy_mcp_handler"
-    )
+    handler_module = importlib.import_module("litellm.responses.mcp.litellm_proxy_mcp_handler")
     monkeypatch.setattr(handler_module, "function_setup", fake_function_setup)
 
     tool_name = "deepwiki-read_wiki_structure"