MCP processes sometime don't killed correctly

[dhnam]

I don't actually know it's problem with agent, or MCP processes, but I'll issue it here.

I'm currently using Windows, newest version of fast agent (until April 18th). (Also, using Gemini as llm.)

I'm using MetaMCP to manage MCPs - it's the only MCP I added to yaml file.

The problem is - even after I turn fast agent off, sometime my memory is full with unkilled MCP processes.

It's not always - sometime it kills every MCP and everything remain fine, but sometime I got full of unkilled processes.

I hope I can find out what exactly is the problem - MetaMCP or fast agent or individual MCPs.

[evalstate]

There seem to be a couple of issues in this area at the moment:
modelcontextprotocol/python-sdk#514
modelcontextprotocol/python-sdk#526

I think narrowing it down may take a while - any input greatly received (I've seen a couple of issues on Windows).

[dhnam]

Oh, so it was the issue of python MCP itself. (Also my docker was overflowing - it was python built on docker)