WebMCP

How does WebMCP work?

A site registers tools, each with a name, a natural-language description, and a schema for its inputs. There are two ways to do this. The imperative API registers tools in JavaScript through document.modelContext.registerTool(), with a name, description, input schema, and an execute function. The declarative API turns an ordinary HTML form into a tool by adding attributes such as toolname and tooldescription.

WebMCP runs inside the browser, in a session where a person has asked an agent to do something. The draft specification says it borrows vocabulary from the Model Context Protocol (tools, schemas, parameters) and is designed for the web platform as a client-side complement to server-side MCP. Tool registration is disabled by default in cross-origin iframes and has to be delegated with the tools permissions policy.

When an agent triggers a declarative form, the browser sets the read-only agentInvoked flag on the SubmitEvent to true, which lets a site count agent submissions separately in analytics. With respondWith(), the page can return a structured result to the agent. With the toolautosubmit attribute, the browser submits the form and navigates as soon as the agent invokes the tool.

Who is developing WebMCP, and what is its status?

The draft is incubated in the W3C Web Machine Learning Community Group, with editors from Microsoft and Google. Chrome offers WebMCP as an origin trial starting with Chrome 149, and developers can test it locally behind a flag. Google’s documentation states that the API is under active discussion and may change.

Example

A support form exposed as a declarative tool:

<form toolname="createSupportRequest"
      tooldescription="Sends a request to customer support."
      action="/support">
  <input name="email" type="email" required>
  <textarea name="message" required></textarea>
  <button type="submit">Send</button>
</form>

From these attributes, an agent can tell what the form does and which fields it needs to fill. Without the optional toolautosubmit attribute, the user still clicks Submit to send it.

FAQ

What is the difference between WebMCP and MCP? MCP connects AI applications to servers, usually outside the browser. WebMCP is a browser API through which the page currently open in a tab offers tools to an agent. Both share terms like tools and input schemas, but WebMCP is a separate specification.

Does WebMCP affect SEO? Google has not documented WebMCP as a ranking factor. It matters for sites where agents are expected to complete tasks, for example bookings, orders, or support requests.

How do I enable WebMCP in Chrome? For local testing, Chrome’s documentation points to the flag chrome://flags/#enable-webmcp-testing. For real visitors, a site has to register for the origin trial.

Last updated: October 8, 2026