Pretty cool.

Just saw a insightful talk https://www.youtube.com/watch?v=eeOANluSqAE

Which basically says "Your API is not an MCP".

The speaker makes some good points, namely that developers try to just expose bare building blocks of API and think that its going to be useful to LLM but turns out they are actually more like humans.

Instead of "building block API", a LLM needs actionable functions that hide some of the complexity from the LLM (like a human).

Further the speaker makes the point that "less is better" - so a list of 75 endpoints if more likely to confuse the API rather than 5 actionable functions.

I'm not commenting this specifically about this project, only saying its changed the way I think about MCPs