What Is WebMCP?

WebMCP is a proposed web standard that lets a website expose structured tools — JavaScript functions or annotated HTML forms — that AI agents can discover and invoke directly in the page.

Instead of reading the DOM and guessing what each button does, an agent gets explicit names, descriptions, and JSON Schemas. The user and agent share the same live page and signed-in session, and tool executions stay visible on screen.

• Typed tools

  One schema-validated tool call replaces a fragile sequence of DOM reads, guesses, and clicks.

• Progressive enhancement

  The human-facing interface stays intact while the page gains an agent-readable layer.

• No user setup

  Tools ship with the website, so visitors do not install or configure them.

Where tools run

MCP server

In a separate server process — local (stdio) or remote (Streamable HTTP)

WebMCP

Inside the web page itself, in the page’s JavaScript context

Who hosts them

MCP server

A developer runs or deploys a server and users configure their client to connect

WebMCP

The website ships them — nothing for the user to install or configure

What they expose

MCP server

Backend capabilities: APIs, databases, files, external services

WebMCP

Client-side capabilities: forms, page state, in-app actions, the signed-in session

Session and context

MCP server

The server manages its own auth and state, separate from any browser session

WebMCP

Shares the live page and the user’s existing signed-in session with the agent

Discovery

MCP server

Client connects to a configured endpoint and lists tools over the protocol

WebMCP

The browser mediates: agents discover tools when the user visits the page

1. 1

   Register

   The page declares a tool with HTML attributes or JavaScript.

2. 2

   Discover

   The browser exposes the tool’s name, description, and input schema.

3. 3

   Execute

   The agent calls the tool and the page handles the action visibly.

<form
  toolname="submit_mcp_server"
  tooldescription="Submit a new MCP server for review"
  toolautosubmit
>
  <input
    name="github_url"
    toolparamdescription="GitHub repository URL"
    required
  />
  <button type="submit">Submit</button>
</form>

await document.modelContext.registerTool({
  name: "submit_mcp_server",
  description: "Submit a new MCP server for review",
  inputSchema: {
    type: "object",
    properties: {
      github_url: {
        type: "string",
        description: "GitHub repository URL",
      },
    },
    required: ["github_url"],
  },
  async execute({ github_url }) {
    const result = await submitServer(github_url)
    return { content: [{ type: "text", text: result.message }] }
  },
})

• Set up Chrome

  Check this browser for WebMCP support, then follow the official Chrome flag setup if the API is not available.

• Use it with ChatGPT

  The ChatGPT desktop app’s built-in browser and ChatGPT Sites support WebMCP tools.

• Inspect registered tools

  The Model Context Tool Inspector extension shows a page’s tools, validates schemas, and lets you call tools manually.

WebMCP is incubated in the W3C Web Machine Learning Community Group, with the goal that any browser with agentic capabilities can implement it.

• Origin isolation keeps a tool’s origin stable for its whole lifetime.
• The tools permissions policy defaults to self, limiting cross-origin frames.
• Tools execute in the page the user is looking at, so actions remain visible.
• Sensitive tools such as purchases can require explicit confirmation.
• Cross-origin discovery is limited to explicitly listed, secure origins.
• Every execution receives an AbortSignal so in-flight work can be cancelled.