Comment by DANmode

MCP apparently reduces probabilistic failures - aka the common fatal flaw of all of these robots (hallucinations, missing stuff in the API doc, etc).

This makes it a little more interesting to me, knowing those results.

It definitely underlines what we already know about the specific weaknesses of LLMs replies/results.