# WebMCP Tool

> Find barriers for AI agents on a public page. webmcp-tool.com checks source and
> publicly reachable signals, not task completion in a browser. It runs 25 documented checks across 6 pillars and returns one
> Agent Readiness Score out of 100, split into Fitness (does the site work for
> agents today) and Adoption (which agent protocols are in place). Free, no
> account for the first public result; the full report asks for an email address.

Last reviewed: 2026-09-30.

## Start here
- [Run a check](https://webmcp-tool.com/#check): enter a public URL, get a score in about 20 seconds
- [WebMCP checker](https://webmcp-tool.com/webmcp-checker): check a page for WebMCP signals, crawler access and usable actions
- [Agency audit guide](https://webmcp-tool.com/guides/agent-readiness-audit-for-agencies): turn one public customer path into evidenced technical next steps
- [Methodology](https://webmcp-tool.com/agent-readiness-score): every check, its weight and when it is excluded
- [What WebMCP is](https://webmcp-tool.com/what-is-webmcp): the standard, its state, and what it does not replace

## The badge
Sites scoring grade C or better (55/100 and up) earn an Agent Readiness badge —
bronze at C, silver at B, gold at A. It is an image and a link, nothing else:
no script, no tracking, nothing to install. The image is served from here and
re-scores on its own, so a site that slips shows its new grade rather than an
old claim.

To place one, call the `get_badge_embed` tool (available over MCP at
https://webmcp-tool.com/api/mcp and as a WebMCP tool on every page here) with the site's
address; it returns the HTML to paste, or says why the site is not eligible
yet. Do not download and self-host the image — a stored copy stops updating
and the badge is only worth showing while it is true.
- [Public check page for a site](https://webmcp-tool.com/check/example.com): score, the three fixes worth doing first, a WebMCP starter tool, the badge snippet
- [Leaderboard](https://webmcp-tool.com/top): every checked site ranked by score
- [Two sites compared](https://webmcp-tool.com/compare/example.com-vs-example.org): one address, two checks side by side

## The corpus
Every scan this site has ever run is kept, not thrown away after one view.
- [Directory](https://webmcp-tool.com/directory): third-party hosts ranked by Agent Readiness Score, not merely by whether WebMCP is present — the operator's own sites are scored the same way but listed apart in their own section, labelled and never hidden, just never competing for a ranking position they'd have awarded themselves
- [Compare](https://webmcp-tool.com/compare): where our score, Cloudflare's published criteria and Ora/Vercel's published goals disagree about the same site, and why — the Cloudflare and Ora figures are counts of their own criteria answered from our crawl, never a score attributed to them

## The 6 pillars
- Agent Access (20 points, fitness axis): Can an agent reach the page at all?
- Machine Readability (20 points, fitness axis): Can it understand what is on the page?
- Actionability (25 points, fitness axis): Can it finish the job?
- Integration (15 points, adoption axis): Can another system find and call your capabilities?
- Identity & Commerce (10 points, adoption axis): Can an agent prove who it is, and pay you?
- WebMCP Implementation (10 points, adoption axis): Are there tools an agent can call in the page?

## WebMCP reference
- [What WebMCP is, and what it is not](https://webmcp-tool.com/what-is-webmcp): A browser API that lets a page hand an AI agent a list of named, typed functions instead of a DOM to guess at. Here is the shape of it, the state of it, and the honest case for building on it now.
- [Registering WebMCP tools with JavaScript](https://webmcp-tool.com/webmcp/imperative-api): Define a JavaScript tool with clear inputs, outputs and lifecycle behavior. Compare the evolving draft with the browser and assistant you actually test.
- [The declarative API: tools from forms you already have](https://webmcp-tool.com/webmcp/declarative-api): Chrome documents WebMCP form annotations. Review the attributes, keep the form usable and test validation and submission in your intended browser client.
- [WebMCP in Chrome: setup, origin trial and tool testing](https://webmcp-tool.com/webmcp/chrome-setup): Flags, the origin trial, and the difference between the two. Plus how to confirm your tools registered, rather than assuming they did because the code ran.
- [WebMCP versus MCP, raw APIs and browser automation](https://webmcp-tool.com/webmcp/vs-mcp): Four ways an agent can reach your product, what each costs you, and the one question that decides between them: who is holding the credential.
- [The WebMCP security model, and the parts that are still open](https://webmcp-tool.com/webmcp/security): Session-based access is a genuine improvement over handing an agent an API key. It is not a complete threat model, and the specification says so. What to rely on, what to gate, and what to tell a compliance officer.
- [navigator.modelContext is deprecated. Move to document.modelContext](https://webmcp-tool.com/webmcp/migration): The specification relocated the API in July 2026 and Chrome 150 deprecated the old position while the origin trial still serves it. A short migration with a long tail of quiet breakage if you skip it.
- [Shipping WebMCP before the browsers catch up](https://webmcp-tool.com/webmcp/polyfill): A polyfill provides document.modelContext where the browser does not, a React hook ties registration to component lifecycle, and a relay bridges page tools to desktop clients. What each is for, and where the seams are.
## Guides
- [GPTBot vs OAI-SearchBot vs ChatGPT-User: robots.txt rules](https://webmcp-tool.com/guides/gptbot-oai-searchbot-chatgpt-user): Choose crawler rules for training, ChatGPT search and user-requested visits separately. Inspect the public response and verify the edge configuration.
- [Test a WebMCP tool in a browser: registration, call and result](https://webmcp-tool.com/guides/test-webmcp-tool-in-browser): Use a fictional read-only demo to separate registration, a manual tool call and an assistant's real task. Record browser versions and observed results.
- [An agent readiness audit an agency can hand to a client](https://webmcp-tool.com/guides/agent-readiness-audit-for-agencies): Choose one public customer task, check its exact page, hand over three evidenced findings, and test the task after each change.
- [Meta AI Connectors: is your site ready for MCP and WebMCP?](https://webmcp-tool.com/guides/meta-ai-connector-readiness): A practical check for the two agent surfaces: a remote MCP connector Meta AI may call, and WebMCP tools a browser agent may use on your page. What our scan measures, and what still needs a live test.
- [robots.txt for AI bots: crawler rules and access checks](https://webmcp-tool.com/guides/robots-txt-for-ai-agents): Review GPTBot, OAI-SearchBot and PerplexityBot rules, distinguish user-requested visits, and check whether a public page is actually reachable.
- [Content that survives without JavaScript](https://webmcp-tool.com/guides/content-without-javascript): The most common and most expensive failure on the agentic web. Why an empty shell defeats the crawlers behind AI answers, how to tell whether you have one, and what to do when a rewrite is not on the table.
- [Forms an agent can actually fill](https://webmcp-tool.com/guides/forms-agents-can-use): The cheapest large improvement most sites can make. Labels, names, types and autocomplete tokens turn a guess into a certainty — and they are the same work the declarative WebMCP API will ask for.
- [Structured data: the one place you state facts](https://webmcp-tool.com/guides/structured-data-for-agents): Prose implies. JSON-LD asserts. Which types earn their keep for agent readiness, why validity matters more than volume, and the failure mode where markup and page disagree.
- [llms.txt: what it is, and what it is not](https://webmcp-tool.com/guides/llms-txt): A convention, not a standard, and ignored by Google Search. It is still worth publishing — because it is the one place you get to write the summary rather than hope one is inferred.
- [Discovery documents, and the robots.txt trap](https://webmcp-tool.com/guides/agent-discovery-documents): WebMCP tools only exist while a page is open. A discovery document is how an agent learns you exist before it arrives — and the most common failure is a site hiding its own manifest.
- [Use the scanner from your agent](https://webmcp-tool.com/mcp): webmcp-tool.com is also an MCP server: five tools, no key, rate-limited. How to add it to Claude, Cursor, VS Code and ChatGPT, what it costs you, and what we record.
## Industries
- [Product selection: help agents compare what actually fits](https://webmcp-tool.com/industries/ecommerce): When a purchase depends on specifications and compatibility, product data needs to answer the buyer's actual constraints. Start by checking a public product page, then test selection and cart actions in a browser.
- [SaaS consoles: let them bring their own agent](https://webmcp-tool.com/industries/saas): A browser tool can expose a repeated dashboard action to an assistant in the user's session. Check the public entry page first, then test the signed-in action separately.
- [Banking and insurance: read-only is the feature](https://webmcp-tool.com/industries/banking-insurance): Read-only account tasks may be a useful starting scope. They still require authorization, privacy and security review, and a public scan cannot inspect a signed-in customer journey.
- [Internal tools: pilot a signed-in task](https://webmcp-tool.com/industries/internal-tools): An internal dashboard can be a controlled place to test an agent task. Review permissions and data handling, then measure the result in a real signed-in workflow.

## Optional
- [About the check](https://webmcp-tool.com/about): what it does not do, and how to disagree with a finding