WebMCP

WebMCP (Web Model Context Protocol) is a proposed browser standard that lets a website tell AI agents which actions they can perform on it, such as searching a catalogue, adding to cart or booking an appointment, and how to call them. Instead of an agent reading screenshots or the page’s HTML and guessing which buttons to press, a WebMCP-enabled page exposes these actions as structured tools with defined inputs and outputs. It borrows its ideas from the Model Context Protocol (MCP), applied inside the browser.

How it works

As of the specification’s state in October 2026:

• The page registers tools through a new browser interface. The draft dated 20 October 2026 exposes it as document.modelContext; earlier proposals and articles call it navigator.modelContext. Each tool has a name, a natural-language description and a schema for its parameters. Chrome describes two routes: a declarative API for standard actions that can be defined in HTML forms, and an imperative API for dynamic actions that need JavaScript.
• A browser-based agent discovers the tools and calls them, much like an LLM’s function calling, while the action runs in the user’s own logged-in session.
• The proposal was put forward jointly by Google and Microsoft and is being developed in the W3C Web Machine Learning Community Group, with editors from both companies. It is a Draft Community Group Report, which is not yet a W3C standard.
• Chrome opened an early preview programme for prototyping in February 2026. It was not yet a finished standard or available by default.

MLforSEO’s AI Search and Agentic SEO course places WebMCP in a layered standards picture: Schema.org provides standardised nouns (what your business is and sells), WebMCP provides standardised verbs at the browse layer, and commerce protocols such as ACP and UCP handle purchases inside AI chats.

How it’s used in SEO and marketing

• Planning for action agents: agents that book, buy or enrol on a user’s behalf look for exposed, machine-executable actions; if none exist, they may fail or move to a competitor that offers them.
• Building the foundation now: real-time inventory and availability data, machine-readable policies and complete structured data are what action tools will connect to.
• Working with developers on which site tasks (search, filter, check availability, book) are worth exposing first.

Example

A clinic’s site registers a check_availability tool (inputs: service, date) and a book_appointment tool. A user asks their browser assistant to book a physiotherapy slot next Tuesday afternoon. The agent calls check_availability, shows the options, and with the user’s confirmation calls book_appointment, without trying to click through the calendar widget.

Limitations and things to watch

• It is an early-stage proposal; the API, names and browser support may change. Check the specification before building.
• The course’s advice is that you don’t need a WebMCP endpoint today, but the entity and attribute layers must be in place first: an agent that cannot verify your entity will not act on your site.
• Exposed actions need the same security, consent and abuse protections as any other interface.

Learn it on MLforSEO

• Your audience is no longer only human
• Agent Readiness Stack Audit
• Agent SEO Implementation Roadmap
• AI Search Optimisation and Agentic SEO