What is Chrome WebMCP?

WebMCP is a proposed web standard being developed by Chrome: a website explicitly declares its capabilities to AI agents. The agent does not have to interpret the interface by its look; the site says what it can do, through JSON Schema definitions of tools and actions.

How is WebMCP different from Schema.org and llms.txt?

They serve different purposes: Schema.org describes who you are (business type, address, contact info). llms.txt suggests what to read about you (key pages and summaries). WebMCP describes what an agent can do on your site (book, search, fill forms). Together they form three layers.

When will Chrome WebMCP be available?

Per Chrome's documentation, there is a flag for local testing (chrome://flags/#enable-webmcp-testing), and the Origin Trial, in which real websites can take part, can be joined from Chrome 149. Full availability depends on the standard being finalized.

Should I implement WebMCP on my website right now?

Not yet. WebMCP is a proposed standard at the Origin Trial stage and may change significantly. For now it makes sense to work on Schema.org and FAQPage markup: stable standards that describe your site for machines. Whether they affect ChatGPT, Gemini, Grok or Claude answers is not publicly confirmed. WebMCP is worth watching, not implementing yet.

May 21, 2026 · Updated 11 October 2026 · AI search

Chrome WebMCP: The Next Layer of AI Visibility After Schema.org

While everyone was reacting to Google I/O, Chrome published documentation for WebMCP, a proposed standard that could affect how AI agents work with websites, and announced an Origin Trial. What WebMCP is, how it relates to Schema.org and llms.txt, and what to do about it.

Three layers that describe a site to AI

Describing a site to AI is becoming layered. Each layer says something different about your site:

Schema.org — Describes who you are: business type, address, contact info, products. The foundation. If you have nothing else, start here.

llms.txt — Suggests what to read about you: which pages matter, in what order. A structured summary at yoursite.com/llms.txt. It is a proposed format, not a standard.

WebMCP — Describes what an agent can do on your site: book, search, fill forms, trigger actions. The new layer Chrome proposes.

What WebMCP actually is

Today, AI agents that interact with websites do so by interpreting the visual interface — reading text, identifying buttons, guessing what clicking something will do. This works, but it's fragile and slow.

WebMCP proposes a different approach: websites explicitly declare their capabilities to AI agents using JSON Schema. Instead of the agent figuring out how your booking form works, your site tells it: here's the form, here are the fields, here's what happens when it's submitted.

Two implementation paths are available: an imperative JavaScript API for complex interactions, and a declarative HTML annotation approach that automatically creates WebMCP tools from existing forms.

What sites can declare with WebMCP

—Booking and reservation forms with structured field mapping
—Product search with filterable parameters
—Support workflows with escalation paths
—Multi-step interactions that require user confirmation
—Backend functions triggered by AI agents on behalf of users

Current status: early

Per Chrome’s documentation (developer.chrome.com/docs/ai/webmcp, updated October 7, 2026), WebMCP can be tested locally behind a Chrome flag (chrome://flags/#enable-webmcp-testing), and the Origin Trial, in which production websites can take part, can be joined from Chrome 149.

The spec is under discussion and may change. It is not a stable standard yet.

What to do right now

Regarding WebMCP specifically, it is optional and experimental, so waiting is reasonable. The standard is not finalized, this is an Origin Trial, and the implementation details will likely change. A WebMCP audit in Lighthouse is not a reason to hurry.

What you can do right now is layers 1 and 2, the foundation AI services can read today:

—Schema.org markup. Describes your business to machines. A necessary baseline, but markup alone does not promise citations: in one lab test (arXiv:2605.25517, 2026; two documents, six models), price, a recent date and topical match mattered more, and formatting-only edits had little effect.
—FAQPage schema. Marks up the questions and answers on a page. An effect on citations is not proven; see the FAQ schema article for details.
—llms.txt. Suggests which pages matter to AI agents. A structured entry point for anyone who reads it; whether major services use the file is not publicly confirmed.

When WebMCP leaves the Origin Trial and stabilizes, we will consider adding it to the Causabi AI score. Until then, it is one to watch.

Check your AI score

The check shows which of the signals (Schema.org, FAQ schema, llms.txt and others) your site has that AI crawlers can read today. It takes about 90 seconds.

Free check: can AI read your site?

See your AI score before you read on

About 90 seconds · No email required · Free