Browser UseTalk to sales

13,352 sites · 45,763 distinct tools · scan finished Oct 19, 2026

Websites with WebMCP tools.

Sites that hand AI agents real tools, like search, add to cart or book, instead of making them click around. We found them by loading 322,714 pages in Chrome on Browser Use Cloud and recording every tool each page registered.

Search the sitesWhat is WebMCP?How we found them
Tranco top 100k sites listed
1.7%
1,270 of the 72,893 that loaded
Distinct tools
45,763
a platform’s set counts once; 112,872 counting every site
Register through Chrome’s API on a re-check
79.3%
317 of 400 random listings, Oct 24, 2026

Kinds of site

All 13,352 sites

Biggest sites first. Search matches tool names, so “checkout” or “book” finds sites that can do it.

13,352 sites

Platforms that ship it for you

Some platforms can add the same tools to the sites built on them. An identical tool set on 8 or more unrelated sites counts as a platform; we list its sites, and the distinct tool count above counts its tools once.

14 more platforms

What is WebMCP?

A proposed web standard for pages to hand tools to AI agents. A page registers each tool with a name, a description and an input schema, and an agent in the browser calls the tool instead of guessing which button to press. The draft spec is a W3C Web Machine Learning Community Group report, not a W3C standard; the explainer covers the why. Chrome offers it as an origin trial from Chrome 149.

There are two ways: JavaScript with document.modelContext.registerTool(), or an HTML form with a toolname attribute. Chrome’s docs cover both.

await document.modelContext.registerTool({  name: "search_products",  description: "Search the catalog. Returns name, price and URL.",  inputSchema: {    type: "object",    properties: { query: { type: "string" } },    required: ["query"],  },  annotations: { readOnlyHint: true },  execute: async ({ query }) => searchCatalog(query),});
app.js

How we found them

  1. 1. We loaded 322,714 pages in Browser Use cloud browsers (Chrome 154): the Tranco top 300k sites (list 467YX), plus candidates from public WebMCP lists and our own search. The scan finished Oct 19, 2026.
  2. 2. Each page got time to load. We recorded every tool it registered: Chrome’s own tool list, Chrome DevTools WebMCP events, and HTML forms with a toolname.
  3. 3. Chrome 154 in our cloud browsers only exposes document.modelContext on sites with a WebMCP origin-trial token. To see tools on every other site, the scanner installed a recorder before the page loaded and logged each tool the page registered. In this scan the recorder also answered navigator.modelContext and provideContext(), from earlier drafts, which Chrome does not offer. So a listing means the page registered tools through one of these or a form when the API was there. In a re-check of 400 random listings on Oct 24, 2026, 317 registered every listed tool through the API Chrome ships, 70 only through the older ones, and 13 only some of them or none. On a site without a token, stock Chrome only offers the API with its WebMCP flag turned on.
  4. 4. A site is listed only if it registered at least one tool when we loaded it. Nothing here comes from a claim we did not see.
  5. 5. Names, categories, summaries and the “ask an agent” lines are written by an AI model from each page’s title, description and tool list; a second pass removed use cases the tools cannot do.
  6. 6. Search engines only index a site page that we loaded again on Oct 24, 2026 and that registered every listed tool through document.modelContext or a form: 253 of the 432 pages that clear our quality bar passed, and 216 are left after removing duplicates.