Agent Readiness Check
webmcp-tool.com
Checked 2026-09-08 06:14 UTC
Agents can partly use this site
An agent can read this site and get some way through it, with friction. The points still on the table are listed below, largest first.
Three fixes worth doing first
Ordered by the points still on the table. Each one comes from a documented check and, where it exists, with the code that fixes it.
01Agent-facing authentication
+4 ptsYou expose machine capabilities but no documented way to authenticate against them.
How to fix it Publish OAuth discovery metadata, or an auth.md describing how machine clients obtain access.
RFC 8414 — OAuth authorization server metadataRFC 9728 — OAuth protected resource metadata
02Declarative tools on forms
+4 pts2 form(s) could be exposed as tools but none is annotated.
How to fix it Annotate the forms you already ship. Because the attribute names are still marked TODO in the explainer, verify them against the current specification before rolling this out widely.
03Errors an agent can read
+3 ptsNothing in the markup describes what a valid submission looks like.
How to fix it Set aria-invalid on the failing field, link the message with aria-describedby, and keep the text next to the field rather than only in a toast.
Your first WebMCP tool, matched to this page
The crawl found no search, shop, booking flow or form on this page, so the example below registers the one tool every site can ship in an hour: a read-only summary an agent can call instead of scraping. Paste it and the page stops being guessed at and starts being called.
registerTool
// Register the first tool an agent can call on this page.
// Works today with the polyfill: npm install @mcp-b/global
await document.modelContext.registerTool({
name: "get_page_summary",
description: "Return a short summary of the current page and links to its main sections.",
inputSchema: { type: "object", properties: {} },
annotations: { readOnlyHint: true },
execute: async (_args, { signal }) => {
const text = document.querySelector("main")?.innerText.slice(0, 2000) ?? document.title;
return { content: [{ type: "text", text }] };
},
});.well-known/webmcp
{
"name": "webmcp-tool.com",
"description": "Return a short summary of this site and links to its main sections.",
"tools": [
{
"name": "get_page_summary",
"description": "Return a short summary of the current page and links to its main sections."
}
],
"pages": [
"https://webmcp-tool.com/"
]
}The manifest at /.well-known/webmcp is the convention ChatGPT Desktop reads to list site tools. It is not part of the score yet; the registered tool above is.
llms.txt
# WebMCP Tool: can an AI agent actually use your website? > Free check across 25 signals: crawler access, machine readability, actionable forms, and the WebMCP structured tools a browser agent can call. One score, the evidence, what to fix first. ## Main pages - [Start](https://webmcp-tool.com/) - [About](https://webmcp-tool.com/about) - [Contact](https://webmcp-tool.com/contact)
Built from this page's own title and description. Replace the three links with your canonical pages.
Put the live badge on the site
The image is served from here and re-scores on its own. Every embed links back to this page.
<a href="https://webmcp-tool.com/check/webmcp-tool.com" target="_blank" rel="noopener">
<img src="https://webmcp-tool.com/badge/webmcp-tool.com.svg"
alt="Agent Readiness Score for webmcp-tool.com: grade B, 85 out of 100 — measured by webmcp-tool.com"
width="208" height="40" loading="lazy" decoding="async">
</a>Agent Readiness Check
The full report, plus a re-check in 30 days
Every check with the evidence it was judged on and the code that fixes it, sent as a link that always re-scans. In 30 days we scan again and tell you what moved.
- All checks, pass by pass, with the exact header, count or path behind each verdict
- Copy-paste fixes generated from your own markup
- A re-check in 30 days with the difference, confirmed by you with one click
One email with the report, one re-check after you confirm it, and nothing else unless you ask. No account. Five reports per company per day. What we do with the address.
Agent Tracking
You know whether agents can read this site. Want to know what they do with it?
One line of script records which assistants send visitors, which pages they fetch, which WebMCP tools they call and whether they finish. No cookies, no personal data, hosted in Germany. The check score shows up in the dashboard and is re-scanned monthly.
<script defer data-domain="webmcp-tool.com" src="https://agenttracking.co/agent.js"></script>
Compare with another site
A competitor, a client, the site you are about to ship. Two checks side by side, on one shareable address.
Open the full reportSee the best-scoring sites
Reports re-scan at most once an hour. Change something and open this address again.
Check another site
One request, about 20 seconds. No sign-up, no email. The result gets a shareable address.