Fix async callable object tools

[stephanlensky]

Makes detection of asynchronous functions more robust in Tool.from_function, allowing usage of asynchronous callable objects, e.g.

class MyAsyncTool:
    def __init__(self):
        self.__name__ = "MyAsyncTool"

    async def __call__(self, x: int) -> int:
        return x * 2

manager = ToolManager()
tool = manager.add_tool(MyAsyncTool())
result = await tool.run({"x": 5})  # 10

Motivation and Context

Fixes #567.

How Has This Been Tested?

Unit tests were added to verify the change.

Breaking Changes

This is a non-breaking change.

Types of changes

• 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

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

Additional context

[mcp-shadow]

Thank you for your contribution!

Suggestion: APPROVE

Comment: This PR correctly addresses issue #567 by enhancing the detection of asynchronous functions in Tool.from_function to properly handle asynchronous callable objects. The implementation follows Python best practices by adding a helper function _is_async_callable that handles both regular async functions and async callable objects with __call__ methods. The changes are minimal, targeted to fix only the specific issue, and include comprehensive tests to verify the functionality. This fix properly supports a valid use case without introducing any breaking changes or performance overhead.

[ihrpr]

Thank you for working on this!

Please can you address limit of loop iterations and it's good to merge.

[ihrpr]

Let's limit loop iterations here

[stephanlensky]

Thank you for the review! What would you suggest? I don't think there is any limit to how many partials you can have applied.

I do want to note that the reference implementation from Starlette is used by every consumer of FastAPI and does not specify a limit.

[stephanlensky]

The CPython standard library has a very similar implementation, also without a limit.

[ihrpr]

scary, but consistent :)

[Kludex]

Starlette also has the same implementation.

[ihrpr]

Thank you!!

[ihrpr]

scary, but consistent :)

[ihrpr]

@stephanlensky please can you merge main and I'll merge!

[stephanlensky]

Should be all set now! Thanks again 🙂