The Lab, in public.

Small working experiments: interfaces that assemble themselves, tools an agent can call, and workflow ideas I want to test before I write about them. Some of them break, which is part of the point. Nothing here is for sale. It is the site showing its work.

See the experiments

Agent surface

Simulated

This site offers three tools to AI agents: search the writing, fetch a post as clean markdown, and send me a message. In a browser that supports WebMCP, an agent can call them directly, and every call shows up in the log below. Everywhere else, the Run buttons call the same code by hand, so you can see exactly what an agent would get.

search_posts
query
get_post
slug
send_message
your emailkindmessage
This sends a real message to me.

Tool calls

No calls yet.

Agents can act. People stay in charge. Sending a message is marked as consequential, so your agent or browser can check with you first, and this site limits how much any one sender can send.

To see it live, open this page in a browser agent that supports WebMCP, or turn on WebMCP testing in Chrome (chrome://flags/#enable-webmcp-testing).

Written up

WebMCP build log

Making this site usable by AI agents (WebMCP build log)

How a static Astro site learned to offer real tools to AI agents through WebMCP, what I had to fix while building it, and the rule I started with that turned out to be wrong.

Oct 5, 2026Text