What is WebMCP? Definition and how it works

WebMCP is a proposed web standard that lets a website register its own actions as structured tools that AI agents in the browser can discover and call.

How it works

Today a browser agent uses a website the way a person does: it reads the page, guesses which button does what, and clicks. WebMCP lets the page say it directly. The site registers tools, each with a name, a plain-language description and a JSON Schema for its inputs, and an agent working in that tab calls the tool instead of simulating clicks.

The draft specification defines a JavaScript entry point, document.modelContext, with a registerTool() method (WebMCP draft). Chrome's documentation also describes a declarative option that turns annotated HTML forms into tools, and notes that both are controlled by a tools permissions policy that defaults to the page's own origin (Chrome docs). A simplified sketch:

// Sketch based on the October 2026 draft. Names may still change.
document.modelContext.registerTool({
  name: "search-docs",
  description: "Search the product documentation and return matching page titles and URLs",
  inputSchema: { type: "object", properties: { query: { type: "string" } }, required: ["query"] },
  async execute({ query }) {
    const res = await fetch(`/api/search?q=${encodeURIComponent(query)}`);
    return JSON.stringify(await res.json());
  },
});

Tools run inside the page with the visitor's existing session, so the agent can only do what that visitor could do.

Status as of October 2026

• The specification is a Draft Community Group Report of the W3C Web Machine Learning Community Group, edited by people from Microsoft and Google. It says itself that it is not a W3C Standard and not on the W3C Standards Track.
• Chrome runs an origin trial starting in Chrome 149 and offers a flag for local testing (Chrome blog). At Google I/O 2026, the Chrome team said Gemini in Chrome will soon support WebMCP APIs (Chrome at I/O 2026).
• Chrome's docs warn that WebMCP is "under active discussion and subject to change in the future."

Why it matters for your website

When an agent can call "check availability" or "request a quote" as a defined tool, fewer tasks fail halfway through a form. WebMCP is a progressive enhancement: people and agents that do not support it see the same page as before.

Because the standard is a draft, treat it as an experiment. Start with one or two high-value actions, keep them read-only or behind your normal confirmation step, and think about prompt injection: a tool should never do more than the visitor could do by hand.

Example

A B2B software site registers two tools: search-docs, which only reads, and request-quote, which opens the site's usual quote form pre-filled and waits for the visitor to submit it. A visitor tells their browser assistant, "Get me a quote for 20 seats from this vendor." The assistant calls request-quote with the seat count, the page shows the filled form, and the visitor reviews and sends it.

Related terms

• Model Context Protocol (MCP): the protocol WebMCP borrows its tool model from.
• Agentic browser: the kind of browser that calls WebMCP tools.
• AI agent: software that chooses and calls tools to reach a goal.
• Prompt injection: why tools need tight limits.
• llms.txt: a read-only way to point AI tools at your key pages.