Did you see token-warden, could be a good match. Be interesting to see how the models compare on the same task especially with different thinking levels...
This is great and definitely under-served right now, will take a look. I have done a lot of primary research on MCP observability so know what I am looking for :)