Repository navigation
FastMCP server death on client HEAD calls #1269
Description
Initial Checks
- I confirm that I'm using the latest version of MCP Python SDK
- I confirm that I searched for my issue in https://github.com/modelcontextprotocol/python-sdk/issues before opening this issue
Description
I am using streamable HTTP for MCP. I see HTTP HEAD calls in the log which killing the server. I expect the server to be resilient to poor user behavior.
My code:
https://github.com/deephaven/deephaven-mcp/tree/main/src/deephaven_mcp/mcp_docs_server
Example logs:
2025-08-13 09:29:26.178 PDT
[2025-08-13 16:29:26,179] INFO: [mcp_docs_server:app_lifespan] MCP docs server ready and yielding context
2025-08-13 09:29:26.179 PDT
INFO: 169.254.169.126:65146 - "HEAD /mcp HTTP/1.1" 405 Method Not Allowed
2025-08-13 09:29:26.179 PDT
[2025-08-13 16:29:26,180] INFO: Terminating session: None
2025-08-13 09:29:26.183 PDT
[2025-08-13 16:29:26,180] ERROR: Error in message router
2025-08-13 09:29:26.183 PDT
Traceback (most recent call last):
File "/app/.venv/lib/python3.12/site-packages/mcp/server/streamable_http.py", line 831, in message_router
async for session_message in write_stream_reader:
File "/app/.venv/lib/python3.12/site-packages/anyio/abc/_streams.py", line 41, in __anext__
return await self.receive()
2025-08-13 09:29:26.183 PDT
^^^^^^^^^^^^^^^^^^^^
2025-08-13 09:29:26.183 PDT
File "/app/.venv/lib/python3.12/site-packages/anyio/streams/memory.py", line 111, in receive
2025-08-13 09:29:26.183 PDT
return self.receive_nowait()
2025-08-13 09:29:26.183 PDT
^^^^^^^^^^^^^^^^^^^^^
2025-08-13 09:29:26.183 PDT
File "/app/.venv/lib/python3.12/site-packages/anyio/streams/memory.py", line 93, in receive_nowait
2025-08-13 09:29:26.183 PDT
raise ClosedResourceError
2025-08-13 09:29:26.183 PDT
anyio.ClosedResourceError
2025-08-13 09:29:26.183 PDT
[2025-08-13 16:29:26,184] INFO: [mcp_docs_server:app_lifespan] Context manager exiting normally
2025-08-13 09:29:26.184 PDT
[2025-08-13 16:29:26,184] INFO: [mcp_docs_server:app_lifespan] Final memory usage: 109.02 MB
2025-08-13 09:29:26.184 PDT
[2025-08-13 16:29:26,184] INFO: [mcp_docs_server:app_lifespan] Final CPU percent: 0.0%
2025-08-13 09:29:26.184 PDT
[2025-08-13 16:29:26,184] INFO: [mcp_docs_server:app_lifespan] Final open file descriptors: 11
2025-08-13 09:29:26.184 PDT
[2025-08-13 16:29:26,185] INFO: [mcp_docs_server:app_lifespan] shutdown active asyncio tasks: 3
2025-08-13 09:29:26.184 PDT
[2025-08-13 16:29:26,185] WARNING: [mcp_docs_server:app_lifespan] Running tasks during shutdown: 3
2025-08-13 09:29:26.184 PDT
[2025-08-13 16:29:26,185] WARNING: [mcp_docs_server:app_lifespan] Running task 1: mcp.server.streamable_http_manager.StreamableHTTPSessionManager._handle_stateless_request.<locals>.run_stateless_server
2025-08-13 09:29:26.184 PDT
[2025-08-13 16:29:26,185] WARNING: [mcp_docs_server:app_lifespan] Running task 2: Task-2
2025-08-13 09:29:26.184 PDT
[2025-08-13 16:29:26,185] WARNING: [mcp_docs_server:app_lifespan] Running task 3: mcp.server.fastmcp.server.FastMCP.run_streamable_http_async
2025-08-13 09:29:26.184 PDT
[2025-08-13 16:29:26,185] INFO: [mcp_docs_server:app_lifespan] shutdown active threads: 1
2025-08-13 09:29:26.185 PDT
[2025-08-13 16:29:26,185] INFO: [mcp_docs_server:app_lifespan] MCP docs server shutting down
My attempts to work around this problem are in this PR. You would need to look at the whole history of what I tried. I included different middleware changes, increasing versions, etc.
deephaven/deephaven-mcp#89
Example Code
My code:
https://github.com/deephaven/deephaven-mcp/tree/main/src/deephaven_mcp/mcp_docs_serverPython & MCP Python SDK
The build can be found at https://github.com/deephaven/deephaven-mcp/blob/main/ops/docker/mcp-docs/Dockerfile. It is python 3.12 with the latest mcp (version >= 1.12.4).
Activity
chipkent commentedon Aug 13, 2025
The ticket for my project that may have more details is: deephaven/deephaven-mcp#90
jzwick-narmi commentedon Sep 15, 2025
I am also having this issue
felixweinberger commentedon Oct 3, 2025
Hi @chipkent thanks for reporting this; would you be able to provide a snippet we could run to isolate the issue you're seeing here? I see you linked to your repo, but it'd be great if you could isolate the repro to the specific issue here if possible.
DinoChiesa commentedon Oct 21, 2025
I observed something related...
HEAD gets a 405 from my FastMCP server, but the server does not "exit" when handling this.
On the other hand, when a faulty client sent a POST with no payload, this caused a parse error in the server, which lead to the same stack trace shown here, and server exit. Correcting my client to send a valid JSONRPC payload allowed the server to stay up.
I agree the server should be able to remain running in the face of ill-behaved clients.
fix: reject unsupported HTTP methods early in session manager
Br1an67 commentedon Mar 1, 2026
I'd like to work on this. The fix rejects unsupported HTTP methods (HEAD, PUT, etc.) early in
StreamableHTTPSessionManager.handle_request()before any transport or background task is created, preventing the ClosedResourceError race condition.PR: #2191
DinoChiesa commentedon Mar 1, 2026
This report and this report both said that the client sees 405 and no server exit when sending the unsupported HEAD method.
What does your fix change?
The latter report states
Does your fix address this problem? rejecting HEAD or PUT before any transport task seems irrelevant to that problem.
What does your fix change?