Free Tool · New
Ranking in AI search only gets an agent to your site. WebMCP is the emerging browser-native standard that lets it click, fill, and submit once it's there. Check three signals in seconds.
Read vs. act
MCP connects an AI backend to your server. WebMCP connects an AI agent to your browser page, while a human is still there.
Three checks, no fluff
Tool manifest, llms.txt, and whether your buttons and inputs expose the accessible names an agent needs to click and fill them.
This is a first pass
A full Lighthouse Agentic Browsing score also checks accessibility tree depth and layout shift. This tool is the fast, honest version of that: three signals, checked instantly.
WebMCP is a browser-native standard, co-developed by Google and Microsoft through the W3C, that lets your website declare a menu of actions an AI agent can call directly: search, book, add to cart, submit a form. Instead of guessing its way through your HTML, the agent reads the menu and calls the function it needs, while a person is still at the keyboard watching it happen.
That last part matters. WebMCP is built for cooperative, human-present browsing, not headless automation. It solves a narrower problem than people assume: not "can an AI find my site" but "once it's here, with a user watching, can it actually do something useful." Ranking in AI search gets an agent to your homepage. WebMCP decides whether that visit ends in a completed task or a stalled one.
It's easy to confuse WebMCP with MCP, the protocol Anthropic built, because the names overlap and both are about giving AI structured access instead of raw scraping. They solve different problems. MCP connects an AI model to your backend through a hosted server: no browser, no human at the keyboard, built for service-to-service work. WebMCP connects an agent to your browser page during a live session. Most businesses will eventually want both. If you haven't built the backend side yet, Faro's MCP Generator creates a working server and the card that describes it in a couple of minutes.
The Declarative API is the fast path: add tool attributes directly to a form you already have. If your search bar, booking flow, or contact form is built from clean, well-labeled HTML, most sites are closer to ready than they think, often with no new backend work at all. The Imperative API is for interactions that live in JavaScript: a product filter, a multi-step checkout, anything dynamic. There, a developer registers a named, typed function through the browser's navigator.modelContext API, and the agent calls it with structured parameters instead of clicking through the UI by trial and error.
Since Lighthouse 13.3.0, Chrome's own auditing tool ships an Agentic Browsing category by default, right next to Performance and SEO. It scores four checks: WebMCP tool registration, accessibility tree health, layout stability (CLS), and llms.txt. Faro's WebMCP Readiness Check covers the two that matter most and can be verified from a server in seconds: your tool manifest and your llms.txt. It also adds a third signal Lighthouse folds into accessibility more broadly: whether your buttons and inputs actually expose the names an agent needs to act on them. Accessibility tree depth and layout shift need a real browser to measure, so for those, run a full Lighthouse audit in Chrome 150 or later.
Discovery, understanding, and action are three separate layers, and most sites are only working on the first one. A site can rank well in ChatGPT or Perplexity's answers and still fail an agent trying to complete a booking on it, because nothing on the page told the agent how. WebMCP is the layer that turns a visit into a completed task. It's also the layer with the least competition right now: almost every business is still fighting over discoverability, and almost none have touched this. Run the full AI Readiness Scan to see all 47 checks across all 7 categories, including where WebMCP sits alongside your llms.txt and structured data scores, or use the Revenue Calculator to see what a low score is actually costing you each month.
Yes. Enter any domain and run the check as many times as you want: no signup, no card, no limit. It's one of Faro's free public tools alongside the llms.txt Generator and robots.txt Analyzer.
Three signals a server can verify in seconds: a WebMCP tool manifest at /.well-known/webmcp.json (or an in-page registration signal), whether llms.txt exists at your root, and whether the buttons and inputs on your homepage expose accessible names an agent can read.
Not always. llms.txt is a text file anyone on your team can write and upload. WebMCP's Declarative API only needs a few attributes added to forms you already have. The Imperative API, for JavaScript-driven actions like a custom checkout flow, does need a developer.
MCP connects an AI model to your backend through a hosted server, no browser involved. WebMCP connects an AI agent to your browser page while a person is at the keyboard. Most businesses will eventually run both. Faro's MCP Generator builds the server side.
No, and it doesn't pretend to. Lighthouse also scores accessibility tree depth and Cumulative Layout Shift, both of which need a real browser to measure. This tool covers the three signals that matter most and are checkable instantly, without installing anything.
What WebMCP is, how it differs from MCP, and what to fix first.
New to this tool? How to use the WebMCP Readiness Check — what it checks, how to read the results, and how to improve your score.