End to end logging

[sheffler]

This addresses Issue #201 and provides a test and a fix.

Motivation and Context

An MCP server may include logging messages (notifications/message) by using the ctx.{info,debug}() call.
Currently it seems that Claude Desktop and Inspector operate properly when an MCP server sends these
notifications, even if those clients do not do anything with them.

In the process of writing a Pure-Python client using the example given in simple-chatbot as a starting point,
if logging notifications are sent, they cause the client to hang.

There is an included fix in the shared/session.py file. I believe the acknowledgement of the received notification is causing a deadlock. Try running the test without the fix to observe server hanging.

The goal is to simply log the received notifications in the Client, and you can see that in the NotificationLoggingClientSession

How Has This Been Tested?

The included test illustrates the pattern.

Breaking Changes

possibly

Types of changes

• [x ] Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to change)
• Documentation update

Checklist

• [x ] I have read the MCP Documentation
• My code follows the repository's style guidelines
• [x ] New and existing tests pass locally
• I have added appropriate error handling
• I have added or updated documentation as needed

Additional context

[dsp-ant]

In gneeral I am not too keen on end to end tests that involve multiple files but are disguised as unit tests. We should see if we can make this test work within one file by using memory streams, etc.

[dsp-ant]

That seems to be a debugging left over?

[sheffler]

These are the lines that are present in shared/session that I believe cause the deadlock. If you uncomment the lines and run the end-to-end test, it will not complete (it will hang). If you comment the lines out, the test runs to completion.

[dsp-ant]

We generally don't use asyncio but anyio.

[sheffler]

Agreed, a more directed unit test might be better going forward after the issue is understood. The scenario in this end-to-end test illustrates the deadlock with enough context to reproduce the issue described in #201 when run.

[dsp-ant]

Agreed, a more directed unit test might be better going forward after the issue is understood. The scenario in this end-to-end test illustrates the deadlock with enough context to reproduce the issue described in #201 when run.

That makes sense. Can we move this over to DRAFT status until it's ready to be properly reviewed?

[sheffler]

This PR should probably be viewed as a bug-recreation rather than code that should be merged. The end-to-end test does a good job at illustrating the deadlock, I believe.