about mcp server restarts and sessions

[luogni]

Hi, I'm developing a mcp server using fastmcp/python and connecting it to librechat. The first conenction works fine, I can use tools and so on, thanks!

Now I have some questions about how to handle:

• mcp server restarts
• tool operations stop

What is supposed to happen when the mcp server restarts while librechat is working? My tools stop working but I'm not really sure if this is a server or client issue. It seems like a client issue.. maybe it's not correctly restarting the session?

What is supposed to happen when I click the "stop/interrupt" button while a tool is running? Right now I see errors on mcp server and then tools stop working in librechat. I tried to replicate with a python test client but I was not able to.

Can you help me understand what's the supposed workflow on these two situations?

mcp configuration:

mcpServers:
  aymcp:
    type: streamable-http
    url: http://aymcp:8000/mcp/
    timeout: 300000
    headers:
      X-User-ID: "{{LIBRECHAT_USER_ID}}"

I see this when I press the stop button:

LibreChat      | 2025-05-15 07:11:08 error: [MCP][User: 681a398295c491234d8e8339][aymcp][slow] Tool call failed This operation was aborted
LibreChat      | 2025-05-15 07:11:08 error: [MCP][User: 681a398295c491234d8e8339][aymcp] Error calling "slow" MCP tool: This operation was aborted
LibreChat      | 2025-05-15 07:11:08 error: [api/server/controllers/agents/client.js #chatCompletion] Tool Error "slow_mcp_aymcp" tool call failed: This operation was aborted
LibreChat      | 2025-05-15 07:11:08 error: [api/server/controllers/agents/client.js #sendCompletion] Operation aborted Aborted

[danny-avila]

These errors are standard for what you are describing, aborting the tool call. You should still be able to continue tool calls, if not, the errors would be different. Is this the case for you?

[luogni]

I'm not able to continue tool calls. I get this error on fastmcp side:

aymcp          | INFO:     192.168.64.7:35134 - "POST /mcp/ HTTP/1.1" 200 OK
aymcp          | SSE response error
aymcp          |   + Exception Group Traceback (most recent call last):
aymcp          |   |   File "/root/.cache/pypoetry/virtualenvs/aymcp-aSnAZHCZ-py3.11/lib/python3.11/site-packages/mcp/server/streamable_http.py", line 511, in _handle_post_request
aymcp          |   |     async with anyio.create_task_group() as tg:
aymcp          |   |   File "/root/.cache/pypoetry/virtualenvs/aymcp-aSnAZHCZ-py3.11/lib/python3.11/site-packages/anyio/_backends/_asyncio.py", line 772, in __aexit__
aymcp          |   |     raise BaseExceptionGroup(
aymcp          |   | ExceptionGroup: unhandled errors in a TaskGroup (1 sub-exception)
aymcp          |   +-+---------------- 1 ----------------
aymcp          |     | Traceback (most recent call last):
aymcp          |     |   File "/root/.cache/pypoetry/virtualenvs/aymcp-aSnAZHCZ-py3.11/lib/python3.11/site-packages/mcp/server/streamable_http.py", line 515, in _handle_post_request
aymcp          |     |     await writer.send(session_message)
aymcp          |     |   File "/root/.cache/pypoetry/virtualenvs/aymcp-aSnAZHCZ-py3.11/lib/python3.11/site-packages/anyio/streams/memory.py", line 242, in send
aymcp          |     |     self.send_nowait(item)
aymcp          |     |   File "/root/.cache/pypoetry/virtualenvs/aymcp-aSnAZHCZ-py3.11/lib/python3.11/site-packages/anyio/streams/memory.py", line 211, in send_nowait
aymcp          |     |     raise ClosedResourceError
aymcp          |     | anyio.ClosedResourceError
aymcp          |     +------------------------------------
aymcp          | ERROR:    ASGI callable returned without completing response.

But even if I restart my mcp server I'm still not able to use mcp tools without restarting librechat.

[danny-avila]

Ok thank you, I'm looking into it.

[danny-avila]

@luogni I was able to reproduce the issue. It seems that disconnects were not being properly cleaned up as expected. I have a fix for this, will ping here when it's merged.

[danny-avila]

Closed by #7400

[luogni]

the "librechat not talking to tools after restart" is fixed.

But I still have issues with the second usage:

• start librechat/mcp tool (streamable http)
• run a slow tool
• stop the tool after some seconds with the "stop button" on librechat interface
• try to run another tool
  At this point I see these errors on fastmcp mcp tool:

aymcp          | INFO:     192.168.64.7:35134 - "POST /mcp/ HTTP/1.1" 200 OK
aymcp          | SSE response error
aymcp          |   + Exception Group Traceback (most recent call last):
aymcp          |   |   File "/root/.cache/pypoetry/virtualenvs/aymcp-aSnAZHCZ-py3.11/lib/python3.11/site-packages/mcp/server/streamable_http.py", line 511, in _handle_post_request
aymcp          |   |     async with anyio.create_task_group() as tg:
aymcp          |   |   File "/root/.cache/pypoetry/virtualenvs/aymcp-aSnAZHCZ-py3.11/lib/python3.11/site-packages/anyio/_backends/_asyncio.py", line 772, in __aexit__
aymcp          |   |     raise BaseExceptionGroup(
aymcp          |   | ExceptionGroup: unhandled errors in a TaskGroup (1 sub-exception)
aymcp          |   +-+---------------- 1 ----------------
aymcp          |     | Traceback (most recent call last):
aymcp          |     |   File "/root/.cache/pypoetry/virtualenvs/aymcp-aSnAZHCZ-py3.11/lib/python3.11/site-packages/mcp/server/streamable_http.py", line 515, in _handle_post_request
aymcp          |     |     await writer.send(session_message)
aymcp          |     |   File "/root/.cache/pypoetry/virtualenvs/aymcp-aSnAZHCZ-py3.11/lib/python3.11/site-packages/anyio/streams/memory.py", line 242, in send
aymcp          |     |     self.send_nowait(item)
aymcp          |     |   File "/root/.cache/pypoetry/virtualenvs/aymcp-aSnAZHCZ-py3.11/lib/python3.11/site-packages/anyio/streams/memory.py", line 211, in send_nowait
aymcp          |     |     raise ClosedResourceError
aymcp          |     | anyio.ClosedResourceError
aymcp          |     +------------------------------------
aymcp          | ERROR:    ASGI callable returned without completing response.

I'm not sure if it's a server or librechat error.. I'm trying to replicate with a python client but I'm not able to do it right now. any idea?

[danny-avila]

Seems like a server issue, could not reproduce.

I tried a few different python MCP servers, including my own mock one: https://github.com/danny-avila/mock_confluence_mcp

mcpServers:
  search-internet:
    command: uvx
    args:
      - duckduckgo-mcp-server
  fetch:
    command: uvx
    args:
      - "mcp-server-fetch"

First cancelled the tool call:

Then, try again