WebMCP is a proposed open web standard that lets websites expose their functionality as structured tools so AI agents can use them reliably — without screen-scraping, guessing, or breaking every time a UI changes. Announced at Google I/O 2026 and now in an experimental origin trial in Chrome 149, WebMCP could fundamentally change how AI assistants interact with the web. Where today’s browser agents click around like a confused intern, WebMCP gives them a clean API to call instead.
Google’s Chrome team moved WebMCP from a behind-a-flag prototype to a public origin trial on May 19, 2026. The standard is developed in the W3C Web Machine Learning Community Group, and the proposal was co-authored by Google and Microsoft — making it a genuine cross-vendor effort, not a Google-only play. Firefox, Safari, and Brave are all potential implementers if the standard advances.
What Problem Does WebMCP Solve?
Today, browser-based AI agents — like Claude’s computer use, ChatGPT Operator, or Google’s own AI Mode in Search — interact with websites by reading the visible DOM and simulating clicks. It works, but it’s fragile. Update a button’s CSS class, restructure a checkout flow, or add a cookie banner, and the agent breaks. Every UI change is a potential failure point.
WebMCP addresses this by letting developers explicitly define what their site can do for an AI agent. Instead of an agent guessing that “the blue button that says Add to Cart” is the way to buy something, the developer exposes a addToCart(productId, quantity) tool. The agent calls the tool. The tool does the thing. No pixel-hunting required.
This is the same idea behind Anthropic’s Model Context Protocol (MCP) for desktop applications — but built for the open web, living in the browser, and governed by an open standards body rather than a single company.
How WebMCP Works: The Two APIs
Developers have two ways to define WebMCP tools for their site:
Imperative API — Write JavaScript to define tools with typed parameters, descriptions, and handler functions. This is the full-featured path for complex actions like multi-step booking flows or structured data queries. You define what the tool does, what arguments it accepts, and what it returns.
Declarative API — Annotate existing HTML forms with WebMCP attributes. If you already have a search form, a checkout form, or a contact form, you add a few attributes and WebMCP treats the form as a tool automatically. No new JavaScript required. This is the low-friction path for getting existing web apps agent-compatible fast.
Both approaches surface the tools to the browser’s WebMCP registry, where agent-enabled browsers (starting with Chrome 149) can discover and call them. Expedia demonstrated a live WebMCP integration at Google I/O, letting an AI agent complete a flight search and booking entirely through defined tools. Shopify and Booking.com are also testing the standard.
WebMCP vs Anthropic MCP: What’s the Difference?
Anthropic’s Model Context Protocol, which powers Claude Code and Claude’s desktop integrations, is a server-to-server protocol — it connects AI models to external tools and data sources via a JSON-RPC layer. It runs outside the browser, typically as a local server or cloud service.
WebMCP is a browser primitive. It lives in the DOM, exposed by websites, and consumed by browser-based AI agents. The two are complementary: an AI assistant might use Anthropic MCP to connect to your calendar and email servers, while simultaneously using WebMCP to interact with web applications inside Chrome. Google has been explicit that WebMCP is designed to be the browser-native complement to server-side MCP implementations.
What It Means for Developers
The Chrome 149 origin trial means developers can start building and testing WebMCP integrations today in Chrome Canary and Dev channels. Google’s documentation at developer.chrome.com/docs/ai/webmcp covers the full API surface.
The practical implication for web developers: if your site does anything useful — booking, search, purchase, data lookup — there is now a formal mechanism to make that functionality available to AI agents without handing over screen control. Companies that implement WebMCP early will have their functionality accessible to AI agents across ChatGPT, Claude, Gemini, and any browser-native agent that adopts the standard.
For an idea of where this goes, compare it to what happened when sites added structured data for rich snippets in 2015. The early adopters got prominent placement in search results. WebMCP is the same bet — except instead of a search snippet, you’re buying a front-row seat in the agentic web. Gemini’s AI Mode in Search, which is already live in 200+ countries, is an obvious early consumer of WebMCP tools at scale.
Browser Support and Timeline
As of June 2026, WebMCP is available as an origin trial in Chrome 149. Origin trials are time-limited experiments that let developers test proposed standards in production sites before they ship as stable features. To participate, developers register at the Chrome Origin Trials dashboard and add the returned trial token to their site’s HTTP headers.
Cross-browser adoption depends on the W3C standardization process. Microsoft co-authored the initial proposal, which makes Edge adoption likely. Mozilla and Apple have not publicly committed. Full multi-browser support is probably 12-18 months out if the standard progresses on a typical W3C timeline.
For teams building AI-native apps or planning AI integrations, the prudent move is to start with the imperative API in Chrome now — both to build the implementation muscle and to have real production data to bring to the standards process.
FAQ: WebMCP Explained
What is WebMCP?
WebMCP is a proposed open web standard that lets website developers expose their site’s functionality as structured tools for browser-based AI agents. Announced at Google I/O 2026 and currently in an origin trial in Chrome 149, it’s designed to make AI-to-website interactions reliable and developer-controlled instead of fragile DOM-scraping.
How is WebMCP different from Anthropic’s MCP?
Anthropic’s Model Context Protocol is a server-to-server protocol for connecting AI models to external tools and data sources. WebMCP is a browser primitive — it lives in the DOM and is consumed by browser-based agents. The two are complementary: MCP connects AI to backend services; WebMCP connects AI to web interfaces inside the browser.
Can I use WebMCP today?
Yes, via the Chrome 149 origin trial. Developers can register at the Chrome Origin Trials dashboard to receive a trial token and test WebMCP on production sites in Chrome Canary and Dev channels. Stable release depends on the W3C standardization process.
Which browsers support WebMCP?
Chrome 149 is the only browser with an active origin trial as of June 2026. Microsoft co-authored the proposal, making Edge adoption likely. Mozilla and Apple have not yet committed to implementation. Cross-browser support is expected within 12-18 months if the standard advances on a typical W3C timeline.
What kinds of websites should implement WebMCP?
Any site with structured, repeatable user actions: e-commerce (add to cart, checkout), travel (search, booking), SaaS (create record, run query), and productivity apps (schedule, submit, file). Expedia, Shopify, and Booking.com are among the early enterprise testers. If your site does something useful, WebMCP lets AI agents do it reliably.
Last Updated: June 2026
