What Is WebMCP? How Websites Expose Tools to AI Agents in the Browser

WebMCP is a proposed web API that lets a website hand AI agents a list of "tools" it can use inside the browser tab. Instead of guessing which button to click, an agent can call a named function such as "add item to cart" or "filter products," and the page runs its own JavaScript to do it.

It is not a finished standard. The current document is a draft from a W3C community group, and browser support is limited to origin trials and experiments. This guide explains who is behind it, how it works, and what a business website should do about it now, checked on 6 November 2026.

What WebMCP actually does

The project's GitHub explainer describes it this way: WebMCP "lets developers expose web application functionality as 'tools'", which are "JavaScript functions with natural language descriptions and structured schemas, designed for AI agent ingestion."

Today, most AI agents that use websites work by looking. They read screenshots, the page code or the accessibility tree, then simulate clicks and typing. That is slow and breaks when a layout changes. WebMCP gives the site a way to say, "Here is what I can do, and here is exactly what input each action needs."

The explainer calls pages that use WebMCP a kind of "in-page MCP server." The tools run in the visitor's own browser session, so they can reuse the page's existing code, login and on-screen state. If an agent cannot finish a task with the tools offered, the explainer says it can "fall back to general-purpose browser automation."

A simple example

The explainer's own example is a to-do app that registers a tool named add-todo. The tool takes one required text field. When an agent calls it, the page runs its normal "add item" code and returns a short confirmation. You can see the exact code in our guide on how to add WebMCP to your website.

The explainer also sketches store scenarios. A dress shop registers tools to fetch products by size and color and to update the product grid. A design site registers an order-prints tool that moves the shopper to the secure checkout page, where the person completes the order.

Who proposed WebMCP and who runs it

The explainer says it was "First published August 13, 2025" by engineers from Microsoft (Brandon Walderman, Leo Lee, Andrew Nolan) and Google (David Bokan, Khushal Sagar, Hannah Van Opstal). It adds that the draft "has evolved significantly, primarily driven by" Dominic Farolino.

The WebMCP specification is labelled a "Draft Community Group Report" of the Web Machine Learning Community Group, dated 2 November 2026. The document itself says it is not a W3C Standard and is not on the W3C Standards Track. Its listed editors are from Microsoft and Google.

That distinction matters. A W3C community group can publish drafts quickly, but a draft report carries no formal standards status. Browser makers decide separately whether to ship it.

Status and browser support (November 2026)

The project keeps an implementation status page. Here is what it and the browser vendors' own pages said when we checked on 6 November 2026.

An origin trial is a time-limited test. Sites sign up for a token, and the feature works only on those sites. Chrome's WebMCP origin trial announcement, dated 9 June 2026, calls origin trials "time-limited programs that offer early access to experimental platform features." The Chrome Status entry lists the trial from milestone 149 through 156 and still shows the feature's overall status as "Proposed."

Chrome's WebMCP developer docs, last updated 7 October 2026, describe WebMCP as a proposed web standard that is "under active discussion and subject to change in the future."

What other browser makers think

Mozilla's standards position is labelled "neutral." The discussion notes a risk that "sites will provide tools that do not match the experience of a user on the page." It also says the name suggests MCP is involved when it is not.

WebKit's position is labelled "oppose." Its concerns cover security, privacy, user consent, accessibility and the choice of venue. The closing comment says "WebMCP proposes a new solution before the actual problem has been established."

So the honest summary is this: two browser engines are testing it, one is neutral, and one is against it.

How WebMCP relates to MCP

MCP (Model Context Protocol) is an open protocol for connecting AI apps to outside tools and data, through servers that AI apps such as Claude and ChatGPT connect to. If you are new to it, start with what MCP is.

WebMCP shares MCP's vocabulary of tools and schemas, but it is a browser API, not a network protocol. The explainer says WebMCP is "designed to complement, not replace, existing backend-focused protocols like MCP." It also explains why the team did not just put MCP in the browser: MCP "lacks native web concepts like origins, standard browser permissions, DOM integration, and tab-level lifecycle management."

Put simply:

• An MCP server runs on your servers. AI apps like Claude or ChatGPT connect to it over the internet, even when no browser is open.
• WebMCP tools live in your web page. They work only while a person has the page open, with a browser agent helping them.

Our WebMCP vs MCP comparison goes deeper into when you need each one.

Safety and consent

The spec's security section covers prompt injection, misleading tool descriptions, privacy leaks and cross-origin rules. By default, tools are available only to the page's own origin and the browser's built-in agent. Access is controlled by a tools Permissions Policy that defaults to the same site.

Chrome's guide to secure tools recommends marking big actions. Its advice: "If executing a tool results in significant actions, such as booking travel or transferring money, set consequentialHint: true so the agent or browser can request user confirmation before execution." The same page warns that "it's impossible to guarantee safety inside of a large language model (LLM)."

Who is already using it

Stripe's docs describe WebMCP tools in Stripe payment pages. Browser agents can read checkout context, select a payment method, fill fields and start payment on Stripe Checkout, Elements and hosted invoice pages. Stripe marks payment submission with consequentialHint and tells agent builders to get the user's confirmation first. Stripe also notes that WebMCP "is an experimental browser capability based on a proposed web standard."

What this means for your website

For most business sites, WebMCP is something to watch and test, not something to rush into production. A sensible plan:

1. Fix the basics first. Clean HTML, clear labels and working forms help every agent, with or without WebMCP.
2. List your key tasks. Search, filter, book, request a quote, add to cart. These are the natural first tools.
3. Prototype on a test page. Chrome's local testing flag lets developers try it without an origin trial.
4. Keep humans in charge of money. Payments and bookings should end with a person confirming.
5. Recheck status every few months. The spec changed as recently as 2 November 2026.

To see how WebMCP fits with other agent standards, read our AI agent protocols comparison or browse the agent protocols hub.

Next step

If you want your site to be ready for browser agents, eSEOspace is a web design agency that can review your key user journeys and prototype WebMCP tools on a test build. See our WebMCP development service for how we scope that work.

Frequently asked questions