WebMCP (Web Model Context Protocol) – a browser API for AI agents and the agentic web

Definition

WebMCP is a W3C proposal and Chrome API that lets websites expose functions as tools that AI agents can call directly in the browser – without a separate server.

What is WebMCP?

WebMCP (Web Model Context Protocol) is a W3C proposal and a JavaScript API that lets websites expose their functions as "tools" that AI agents can call directly in the browser. The standard is being developed by the W3C Web Machine Learning Community Group and has been available since February 2026 as an early preview in Chrome.

Technically, the website's JavaScript registers functions – for example add-to-basket, book-appointment or search-products – through an API that the browser exposes to installed AI agents. The agent can then call the function directly without interpreting HTML or simulating clicks.

WebMCP vs MCP – what is the difference?

Both standards carry the name "Model Context Protocol", but they solve different problems:

• MCP is a protocol that connects an AI model to external tools via a server (developed by Anthropic in 2024).
• WebMCP runs in the browser and connects the website's own functions directly to the AI agent – without a separate server and without data leaving the device.

Think of it like this: MCP is how an AI agent talks to Slack or GitHub. WebMCP is how an AI agent talks to your website, while the user has the website open in their browser.

WebMCP vs schema markup – from a descriptive to an actionable web

WebMCP is often compared with schema markup, but the comparison is not entirely fair.

• Schema markup describes what a piece of content is (a product, an event, a review).
• WebMCP describes what the visitor can do on the page (buy, book, filter, submit a request).

If structured data made the web understandable to search engines, WebMCP makes the web usable for AI agents. Schema is descriptive; WebMCP is functional.

How does WebMCP work technically?

A website registers tools via a JavaScript API exposed by the browser. Each tool is defined by:

• Name – for example placeOrder or bookAppointment
• Parameters – which inputs are required
• Description – a plain-language description so the AI agent understands what the tool is for
• Function – the actual JavaScript code that runs when the agent calls the tool

When a user asks their AI agent (in Chrome or via an installed extension) to carry out an action on the website, the agent finds the available tools, chooses the right one and calls it. The whole exchange happens locally in the browser – the website's own code runs – so authentication, sessions and payment details work exactly as they do when a person uses the website.

What does WebMCP mean for SEO and AEO?

WebMCP introduces a new dimension of optimisation that is closer to AEO than to traditional SEO. As AI agents become the first interface many users encounter, it is no longer enough to show up in an answer – the website has to be usable.

AI has already fundamentally changed how search and visibility work, as we explain in detail in How is SEO affected by AI?. WebMCP is the next layer: the point is not just to show up in an AI answer but to be the website the agent can actually act on for the user.

Concrete implications:

• Visibility no longer always leads to a click. An agent that can book, buy or filter directly reduces page views but increases completed deals.
• Technical implementation becomes a competitive advantage. The first websites to expose their core flows via WebMCP will be preferred by agents in the same way that mobile-friendly websites were preferred in 2015.
• Measurement changes. Classic KPIs such as CTR and sessions become less relevant. Tool calls, agent-driven conversion and brand mentions in AI answers take over.

For e-commerce, booking websites and SaaS companies with clear conversion flows, this is not a nice-to-have but the next step in optimisation.

WebMCP and the agentic web

In February 2026, Forbes described WebMCP as the browser-based backbone for the agentic web. That captures why the standard matters: without it, every AI agent has to simulate human interaction on every website, which is fragile and raises security problems.

Where AI Overviews changed how companies show up in AI-generated search answers – an area we cover in AI Overviews and GEO – WebMCP takes the next step: from cited to called. The AI agent chooses not just which website to mention in an answer but which website to actually carry out an action on.

With WebMCP, AI agents get a structured interface to the web, in the same way that REST APIs brought structure to data sharing in the 2010s. The difference is that WebMCP is designed specifically to turn natural language into action.

For companies that want to build agent-ready experiences, we work closely with our sister brand AI Partner, which specialises in AI agents, RAG systems and MCP integrations.

Browser support and current status

WebMCP is still at an early stage – but no longer hypothetical. Status in 2026:

• Chrome: Early preview available since February 2026, enabled via a flag.
• Edge: Expected to follow the Chromium implementation.
• Safari and Firefox: No official word.
• Specification: Active development in the W3C repo on GitHub.

We expect the standard to move from preview to stable in 2026–2027. Companies that want to get ahead should start mapping their core flows and deciding which functions to expose as tools – just as they once mapped which pages needed schema markup.