I donβt quite understand. Are you allowing some AI to run wild with Emacs?
Do you have your emacs config online? I'm struggling to get mcp.el and gpt.el to work together.
So let' me expand the idea I little as I'm too close to this. When you add an elisp-eval tool you give the llm a space to iterate and try out code, and in this case you can have it iterate in the repl on code that produces the behavior you'd want from a gptel based llm tool (ie (edit-buffer ...) (save-buffer...)) Then you can have it generate code into a library of gptel tools that allow it to use emacs. But yes in response to your question, you would be letting the LLM eval elisp... If you want to spin it up in a container or use sandbox-exec you can of course.
@livingfossil my mcp.el integration broke bc both tools are moving so fast adn I haven't fixed it myself yet.. bc I'm using claude desktop most of the time.
yeap yeap, sounds about right. I'm trying to get work to pay for claude. The free tier couldn't get very far with clojure-mcp before running up against length limits.
Interesting π€