Fix dynamic tooling

[kyoncal]

Summary

Changes

• Removed mcp_proxy_manager.py and redundant MCP server.
• Moved filtering to middleware
• Moved middleware adding to server.py
• Fixed and added relevant unit tests
• Fixed logging by outputting to the correct channel

User experience

The user's tools should pass through the proxy transparently instead of being incorrectly cached.
The user should be able to see proper logs

Checklist

If your change doesn't seem to apply, please leave them unchecked.

• I have reviewed the contributing guidelines
• I have performed a self-review of this change
• Changes have been tested

• These changes were tested manually, alongside the suite of automated tests. The manual tests consisted of creating a simple client to repeated query a mock-server through the proxy to verify that the proxy works and now has the desired behavior of querying the back end instead of caching results.

• Changes are documented

Is this a breaking change? (Y/N)

• Yes
• No

Acknowledgment

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.

[arangatang]

Thanks for the changes, one major thing, we need to ensure that proper logging is maintained, did you verify that errors/info/debug logs are propagated properly to i.e. qcli, cline, cursor, terminal etc?
A user must be able to identify problems and such and last I remember from the terminal logs were not showing up. At the very least i'm thinking we should have smth like:

1. Connection to endpoint successfull
2. Exceptions bubbling up when stuff breaks like network
3. Optional - ListTools called (and how many was filtered out)

[wzxxing]

Please also add some description of how this change was tested.

[wzxxing]

You can run pre-commit install so that the linters are checked before updating the PR.

[kyoncal]

Thanks for the changes, one major thing, we need to ensure that proper logging is maintained, did you verify that errors/info/debug logs are propagated properly to i.e. qcli, cline, cursor, terminal etc? A user must be able to identify problems and such and last I remember from the terminal logs were not showing up. At the very least i'm thinking we should have smth like:

1. Connection to endpoint successfull

2. Exceptions bubbling up when stuff breaks like network

3. Optional - ListTools called (and how many was filtered out)

Thanks for the comment. The logs have stayed essentially identical, outside of the change in behavior with regard to removing the mcp_proxy_manager. Also the logging has been fixed, so the user can now see logs in their terminal. I believe this addresses your points. Let me know!

[wzxxing]

why is this run_async lifted to this function?

[kyoncal]

I directly replaced the first call to mcp_proxy_manager with it, although it's original placement outside of this function does make more sense and we can directly run the function.

[wzxxing]

local_mcp is not used any more, can we directly run this function? Since the fastmcp instance created in main is not used anymore?

[kyoncal]

We'd have to use asyncio.run somewhere, since we can't really await the main method, so we'd just be hoisting it elsewhere. I think it makes sense where it is.

[wzxxing]

No, I meant the local_mcp is not used any more, we can remove this argument, also remove the FastMCP instance created in main.

[arangatang]

Rant for another PR/issue (we should review the code for this in general).

1. We should not have default values where they are not needed, we should fail rather than accidentally set defaults if omitted. Makes stuff less prone of unexpected behaviour as the code grows.
2. Ideally we should probably be using kwargs instead of args everywhere, more verbose but much more scalable.
3. We should use proper typing

Probs I should create an issue for this tbh, when thinking about it.

[wzxxing]

This is not used anymore.

We can rename setup_mcp_mode to run_mcp_proxy

[wzxxing]

Then simply do asyncio.run(run_mcp_proxy(args)) here