Proactor Windows warning on cleanup when using stdio_client

[S4mpl3r]

Describe the bug
When connecting to a sdio MCP server on Windows, a warning is printed during the final cleanup. This does not crash the program but indicates that there might be a need for a more graceful shutdown of subprocess transports.

To Reproduce
Run the following script:

import asyncio

from mcp import ClientSession, StdioServerParameters, stdio_client


async def main():
    server_params = StdioServerParameters(
        command="npx", args=["-y", "@playwright/mcp@latest"]
    ) # Or any other server

    async with stdio_client(server_params) as (read, write):
        async with ClientSession(read, write) as session:
            await session.initialize()
            print(await session.list_tools())
            await asyncio.sleep(2.0)


if __name__ == "__main__":
    asyncio.run(main())

Running the above, we get a bunch of warnings and error messages at the end:

Exception ignored in: <function BaseSubprocessTransport.__del__ at 0x...>
Traceback (most recent call last):
  ...
ValueError: I/O operation on closed pipe

Expected behavior
No warnings or errors should be printed during final cleanup. All subprocess transports and associated resources should be closed cleanly without attempting operations on closed pipes.

Desktop:

• OS: Windows 10
• Version: Python 3.11 (mcp v1.6.0)

[t94j0]

I'm running into the same thing. From my attempt to debug, it looks like the finally branch of the app lifespan is never called

[ashrobertsdragon]

This is an old asyncio bug with the Windows Proactor event loop https://github.com/python/cpython/issues/83413#issue-1199028070. They thought they fixed it a couple years ago, but apparently there are a still a few cases where it still slips through. I can confirm it happening with Windows 11 and Python 3.13 as well.
Suppressing the ValueError when closing the session seems to do the trick. So, for your example, it would probably look like this:

import asyncio
from contextlib import supress

from mcp import ClientSession, StdioServerParameters, stdio_client


async def main():
    server_params = StdioServerParameters(
        command="npx", args=["-y", "@playwright/mcp@latest"]
    ) # Or any other server

    with supress(ValueError):
        async with stdio_client(server_params) as (read, write):
            async with ClientSession(read, write) as session:
                await session.initialize()
                print(await session.list_tools())
                await asyncio.sleep(2.0)


if __name__ == "__main__":
    asyncio.run(main())```

[RayL0221]

I have also encountered the same error. Is there a solution for this issue yet? My environment: Windows 11 + Python 3.13.

[DataXujing]

the same error

[Jeomon]

same error am also stuck at when I created an MCP

[hkbb2014]

same here
python 3.13.1
v1.6.0

[hkbb2014]

I tried mcp 1.5.0, it is working fine.
I don't know why 1.6.0 has such problem.

[ashrobertsdragon]

I tried mcp 1.5.0, it is working fine. I don't know why 1.6.0 has such problem.

1.5.0 left zombies that were resistant to cleanup. The fix in 1.6.0 resolved that issue, but created a race condition that would result in this error. It took me a while, but I'm pretty sure I found a workable solution that results in successful cleanup 100% of the time without race conditions and have submitted a pull request for it last night.