AI agents that browse the web on someone's behalf have a hard job. They look at a page, guess which box is the email field, guess which button submits, and hope nothing moved.

WebMCP is Chrome's answer to that. Instead of guessing, an agent can ask the page: what can you do? The page answers with a list of tools, each with a name, a description and a schema for its inputs.

It's still experimental, but it's already in use on more sites than I expected.

How many sites already use it

Browser Use ran a crawl to find out. According to their WebMCP directory , they loaded 322,714 pages in Chrome and found 13,352 sites exposing 45,763 distinct tools .…