WebMCP (Web Model Context Protocol)

WebMCP (Web Model Context Protocol) is a proposed web standard that lets a website declare its own actions, such as filling a form or adding a product to a cart, as structured tools an AI agent can discover and call directly inside the page. The proposal is incubated at the W3C Web Machine Learning Community Group, with the goal that any agentic browser could implement it. It is available as an experimental preview in Chrome, first behind a developer flag starting with version 146, then as a wider origin trial starting with version 149, with further changes expected as the standard evolves.

On the technical side, developers have two options. The declarative API adds HTML attributes directly to an existing form (toolname, tooldescription, toolparamdescription on each field): the browser then automatically builds the tool definition and its JSON Schema. The imperative API uses a JavaScript call such as navigator.modelContext.registerTool(), with a name, a description, an input schema and an execute function, for cases a plain form cannot express. Either way, no server is required: tools run inside the page's own JavaScript, within the user's already signed-in session.

WebMCP differs from MCP (Model Context Protocol) in the layer at which it operates. MCP exposes server-side capabilities, such as a database or an internal tool, through a separate server process that an AI client configures and connects to. WebMCP instead exposes a site's client-side actions, such as a form or a page function, directly inside the page the person is already viewing, discovered by the browser at the moment of the visit rather than through manual client configuration. The rule of thumb its proponents suggest: MCP for backend capabilities, WebMCP for actions inside a page the person is already using.

For BotRank, the stakes go beyond the technical novelty. As agentic browsers and AI shopping assistants increasingly act directly inside pages rather than only reading them, exposing clear, well described WebMCP tools could become a visibility factor alongside structured data and crawlable content, much like how it already works for AI search citation today. See the Computer Use and Agent2Agent Protocol (A2A) glossary entries, and BotRank's technical audits feature.

An ecommerce site adds declarative toolname="add_to_cart" and tooldescription attributes to its product form. A shopping agent can then add the item to the cart in a single structured call instead of parsing the page and simulating a click. The action still happens inside the user's own signed-in session.

A B2B software company registers a "request_quote" tool through navigator.modelContext.registerTool(). An AI agent helping a buyer compare vendors can then submit a structured quote request without the company having to build and maintain a separate public API.