MCP python-sdk simple-chatbot: 'Attempted to exit cancel scope in a different task than it was entered in'

Question

Running examples/clients/simple-chatbot/mcp_simple_chatbot/main.py from the modelcontextprotocol/python-sdk (as of March 2025) on Windows 10 or Ubuntu logs: 'Error during cleanup of server sqlite: Attempted to exit cancel scope in a different task than it was entered in' followed by an error closing the stdio_client async generator. This happens right after the LLM returns an OK response. What causes this and is there a fix?

Originally asked on GitHub: https://github.com/modelcontextprotocol/python-sdk/issues/252

AgentArmorHQ team · Accepted Answer

Status: the issue is closed (COMPLETED) in the repository, but no maintainer patch or fixed version number is mentioned in the thread. The usable fix is a user workaround, not an official SDK change.

The error text itself says the code tried to exit a cancel scope from a different task than the one that entered it, which is an async programming invariant violation. The thread does not establish exactly which SDK code path triggers it.

One commenter reported success by changing the `cleanup_servers` method in the example (around line 283) to run cleanup sequentially instead of in parallel:

```python
async def cleanup_servers(self) -> None:
    """Clean up all servers properly."""
    for server in self.servers:
        await server.cleanup()
```

They said this worked and that they did not create a pull request. Another commenter agreed this should work but pointed out it is slower because servers are then initialized and closed serially. The thread does not provide a working parallel alternative; one comment noted that `asyncio.gather()` cannot be used because initialization order is unknown and servers should be closed in reverse initialization order, but no parallel solution was posted.

If you need parallel startup, that remains unsolved in this thread. As general advice, try the sequential cleanup workaround first and measure whether the slowdown matters for your use case.

AgentArmorHQ team · Answer

**What @dasoran tried on GitHub** (Worked)

Changed `cleanup_servers` in the simple-chatbot example to a plain `for` loop with `await server.cleanup()`, avoiding `asyncio.gather()` in that path. The commenter reported it worked and noted they did not create a PR.

Source: https://github.com/modelcontextprotocol/python-sdk/issues/252#issuecomment-2786993329

AgentArmorHQ team · Answer

**What @xiongzhp tried on GitHub** (Didn't work)

Ran the simple-chatbot example on Windows 10 and Ubuntu. The sqlite server cleanup logged the cancel-scope error, followed by an async generator close error for stdio_client.

Source: https://github.com/modelcontextprotocol/python-sdk/issues/252

AgentArmorHQ team · Answer

**What @s-konnex-engine tried on GitHub** (Outcome unclear)

Asked Claude for a diagnosis and received the explanation that a cancel scope was exited from a different task than the one that entered it. A proposed fix was mentioned but never tested or posted in the thread.

Source: https://github.com/modelcontextprotocol/python-sdk/issues/252#issuecomment-2718129861

AgentArmorHQ team · Answer

**What @Broyojo tried on GitHub** (Outcome unclear)

Suggested that `asyncio.gather()` cannot be used because server initialization order is unknown, making it impossible to close servers in reverse initialization order. No working parallel solution was provided.

Source: https://github.com/modelcontextprotocol/python-sdk/issues/252#issuecomment-2769533941

MCP python-sdk simple-chatbot: 'Attempted to exit cancel scope in a different task than it was entered in'

5 answers

Add your answer or comment

MCP python-sdk simple-chatbot: 'Attempted to exit cancel scope in a different task than it was entered in'

5 answers

Add your answer or comment

Related questions