Client hangs when ctx.info() logging with pure Python client

[sheffler]

Describe the bug
When a tool includes client-side logging with, the client seems to hang.
The following tool works ok with Claude Desktop and Inspector, but does not complete with a Python client.

@mcp.tool()
async def simple_tool_with_logging(x:float, y:float, ctx:Context) -> str:
    await ctx.info(f"Processing Simple Tool")
    logger.debug("IN SIMPLE_TOOL")
    await ctx.report_progress(1, 2)
    return x*y

To Reproduce
Will check in a PR with test and proposed fix.

Expected behavior
I would expect the tool to run to completion.

Screenshots
If applicable, add screenshots to help explain your problem.

Desktop (please complete the following information):

• OS: OSX
• Browser [e.g. chrome, safari]
• Version [e.g. 22]

Smartphone (please complete the following information):

• Device: [e.g. iPhone6]
• OS: [e.g. iOS8.1]
• Browser [e.g. stock browser, safari]
• Version [e.g. 22]

Additional context
Add any other context about the problem here.

[samuelcolvin]

I've spent most of my day held up by this bug, see pydantic/pydantic-ai#1140 (comment)

On main the problem is at

python-sdk/src/mcp/shared/session.py

Lines 344 to 346 in c897868

	await self._incoming_message_stream_writer.send(
	notification
	)

If I change these lines to

if not isinstance(notification.root, LoggingMessageNotification):
    await self._incoming_message_stream_writer.send(
        notification
    )

Everything works correctly, obviously that's not a proper solution, but it demonstrates the problem.

In case it helps, to reproduce this problem, run

npx @pydantic/mcp-run-python sse

(see pydantic/pydantic-ai#1140 for context)

Then run

import logging
from logging import basicConfig

from mcp import ClientSession, StdioServerParameters
from mcp.client.sse import sse_client
from mcp.client.stdio import stdio_client
from devtools import debug

import logfire


logfire.configure()
logfire.instrument_httpx(capture_all=True)
basicConfig(handlers=[logfire.LogfireLoggingHandler()], level=logging.DEBUG)
# disable httpx logs
logging.getLogger('httpcore.http11').setLevel(logging.CRITICAL)
logging.getLogger('httpx').setLevel(logging.CRITICAL)
logging.getLogger('urllib3.connectionpool').setLevel(logging.CRITICAL)

code = """
import numpy
a = numpy.array([1, 2, 3])
print(a)
a
"""


async def call_tools(session: ClientSession):
    await session.initialize()
    await session.set_logging_level('debug')
    tools = await session.list_tools()
    debug(tools)
    result = await session.call_tool('run_python_code', {'python_code': code})
    debug(result)


async def sse():
    async with sse_client('http://localhost:3001/sse') as (read, write):
        async with ClientSession(read, write) as session:
            await call_tools(session)


async def stdio():
    server_params = StdioServerParameters(command='npx', args=['@pydantic/mcp-run-python', 'stdio'])
    async with stdio_client(server_params) as (read, write):
        async with ClientSession(read, write) as session:
            await call_tools(session)


if __name__ == '__main__':
    import asyncio

    with logfire.span('running sse'):
        asyncio.run(sse())

Obviously you can disable the logfire bit, I was using it to see the raw http requests, I see the same hanging behaviour with both SSE and stdio.

As I mentioned in the comment linked above, you can also comment out await session.set_logging_level('debug') to avoid the issue.

[sheffler]

Those are the same three lines i identified, and I believe is deadlock. PR #202 is an end-to-end test that illustrates the condition.