Getting MCP

Nice bit of blog metab­o­liza­tion here: I read Matt Webb’s post about the Model Con­text Protocol with interest, but didn’t really “get it”. Chase McCoy read it, got it, wrote about it — and his post, I get!

Thinking together, right?

Anyway, having now gotten it — sort of — I under­stand that there’s a deep brewing poten­tial here, a cre­ative space opened by the question, “What kinds of strange tools might I pro­vide to a reasons-ing lan­guage model?”

The stranger the better, obviously. The canon­ical “agentic AI” exam­ples of “browse the web for me” and “book a trip for me” are so dull, so deadly.

I don’t believe any of the models presently have a way to explic­itly wait—to say, okay, I am going to apply this tool, then relax for … an hour? a week? two months? until I learn the result, then proceed. It seems like that could pro­duce some inter­esting effects. For some reason I am imag­ining a bonsai bot.

Or maybe “waiting” will actu­ally be a capa­bility accessed through an MCP server? Just think, you could be the one to pro­vi­sion Claude with the most pow­erful tool of all: patience.