Pretty cool. Any thoughts on adding a human-in-the-loop approval button on certain MCP calls?

I imagine sometimes I might want to require approval for certain tool calls all the time, whereas other times I might just want to add that to the prompt.

Yes, this is something you'll be able to implement the middleware functionality that's on our roadmap.

MCP also has an upcoming feature called Elicitation that you might find interesting: https://modelcontextprotocol.io/specification/draft/client/e...