Websites Can Now Expose Page-Scoped Tools to AI Agents via WebMCP

Browsers are gaining a native way for websites to publish tools through document.modelContext, a proposed WebMCP interface, so agents can read a page's available actions. The approach treats the page itself as a tool registry that agents can browse while a human stays in control. The author's implementation exposes 16 tools across the site, with focused subsets scoped to individual pages. Agents can submit a request, but answers come asynchronously from a human operator, and a consent setting controls whether answers become public. The article also describes production bugs caught during verification, including a stale status response and a caching problem.
This is an AI-generated summary. ShortSingh links to the original source for the complete article.


Discussion (0)
Log in to join the discussion and vote.
Log in