Free AI agent readiness and MCP check

Type your address. A real crawl of up to 25 pages tells you in one sentence what AI agents can reach, read and use, whether they can find an MCP server, and the next changes to make.

Free · no account · we obey your robots.txt

Free tool · no account

Example result · a fictional shop, not your site

24 of 43 checks passed 0 to fix · 12 not set up · 7 recorded · 60 not checked on this scan

Your next changes

  1. Say in robots.txt how AI may use your content
  2. Describe your business in JSON-LD
  3. If you run an MCP server, point to it from /.well-known/ai-catalog.json

Your own result opens with one sentence on what AI agents can reach, read and use, and each change comes with what to paste, at a link you can keep.

How this free agent readiness check works

  1. Type your website address and press Check my site. No account and no email are needed to start.

  2. Watch the crawler open your pages as it goes. You can close the tab: your result will be at its link.

  3. Read what AI agents can reach, read and use, how many checks passed, and the next changes to make.

  4. Keep the link, or share it with whoever fixes your site.

What the check looks at

The same crawler that runs inside Convertlyft, pointed at your address, reading what an AI agent without a browser reads. It is ConvertlyftBot: what it reads and how to slow it down or block it.

  1. Your robots.txt, for each AI agent on our dated list, grouped by what the agent is for: training, search, or opening a page somebody asked about. Then a page of your site, asked for as each agent your robots.txt lets in, to see what your server actually does.
  2. Your llms.txt, a markdown copy of a page, and the agent files at well-known addresses, such as an A2A agent card, an MCP server card and an API catalog. Each is asked for once and read as a document; nothing is ever posted to it.
  3. Your HTML: the WebMCP tool forms it declares and what is wrong with them, your structured data, your labels.

What you get, in this order

  1. One sentence on what AI agents can reach, read and use on your site
  2. How many checks passed out of how many ran
  3. The next changes to make, each with what to paste where the standard fixes its shape
  4. Whether an AI agent can find an MCP server
  5. The state of each of the 103 checks, including the ones we do not run yet, each marked as such

MCP readiness check

An MCP server lets an AI agent use your service directly. The check asks your site for the file that tells an agent where one is: the MCP catalog file at /.well-known/ai-catalog.json, and up to 3 server cards it names. Each is read as a document, and the result says whether the catalog file is published and whether it is a valid catalog.

It does not connect to an MCP server. Whether the server answers when asked, its sign-in details and its proof on the MCP Registry are listed in the result as not checked on this scan. If you run an MCP server and no catalog file points to it, the next change reads: “If you run an MCP server, point to it from /.well-known/ai-catalog.json”.

What it checks in your llms.txt

The check asks for /llms.txt once. If the address answers “not found”, the result says the file is not published. If there is a file, the check runs the test Chrome’s Lighthouse runs on it, which looks for three things:

  • a title line, such as # Your site name;
  • at least one link, written as [text](address);
  • at least 50 characters in the file.

A server error at that address (an answer of 500 or above) counts as an error, whether or not the site means to publish the file. That is all this check reads in the file: it does not test it against the whole llms.txt proposal. Search engines do not use this file; some AI tools read it.

What the agent levels mean

  • L0 Unreachable: robots.txt shuts agents out, cannot be read, or something in front of the site refuses an agent robots.txt allows.
  • L1 Reachable: robots.txt was read and lets agents in.
  • L2 Legible: reachable, and no page the crawl read is an empty shell filled in by JavaScript.
  • L3 Operable: legible, and the site’s tool forms have no error, or every page loaded in a browser passes the agent-accessibility check.
  • L4 Agent-native: operable, WebMCP tools are active in Chrome on at least one page, and no error was found in sections 3 to 7 of the checks.

The free scan reads the HTML your server sends and never loads a page in a browser, so it can place a site from L0 to L3. L4 needs your pages loaded in Chrome, which an account’s crawl does. If an earlier Chrome check of your site is on record, the result names it with its date: that reading is from that day, not from this scan.

Agent readiness or WebMCP?

They are the same crawl, read two ways: one scan’s result opens on either page. Start here for the whole site. Use the WebMCP checker when you are building tools for an agent in Chrome.

CompareAgent readiness checkWebMCP checker
What it answersWhat AI agents can reach, read and use on your siteThe WebMCP tools your HTML declares, and what an AI agent in Chrome would get
Result opens onOne sentence, checks passed, your next changeYour tool forms, page by page, against a dated revision of the draft standard
Also showsWhether an agent can find an MCP server, the agent levels, all 103 checksA plain-English guide to WebMCP and how to enable it in Chrome

Neither one calls a tool or submits a form, and neither loads a page in a browser. Whether Chrome actually holds your tools needs an account’s crawl, and that is what L4 asks for.

What it will not do

Call a tool, submit a form, or claim anything it did not measure. A check that could not look says so rather than passing.

The crawler’s level for your site, the health of each part you have adopted and the formula behind it are under How this was measured. A standard you have not adopted is shown, never scored.

Questions about the check

Is the check free, and do I need an account?

It is free and needs no account. Type your address and the crawl starts; the result comes with a link you can keep.

Does it check my whole site or one page?

Up to 25 pages. An account crawls much more of your site and keeps the to-dos in one place.

Does the crawler obey robots.txt?

Yes. It obeys your robots.txt and identifies itself in your logs, except the requests sent under each AI agent’s name, which is what this check measures.

What if I scan the same site again?

If we scanned it recently, you get that result back rather than a fresh crawl of your server.

Is WebMCP the same as an MCP server?

No. An MCP server is a separate service an AI app connects to. WebMCP tools live in your page and run in the visitor’s own browser tab. This check looks for both.

A scan tells you what agents find. Convertlyft keeps watching.

Far more of your site

The free scan opens up to 25 pages. An account crawls much more of your site and keeps the to-dos in one place.

The browser’s side

Whether Chrome holds your tools, what each one asks for, and what an agent reading the page’s accessibility tree gets.

What to do: one change and the evidence

Every to-do in the app comes with its numbers and a Show me that opens them.

Your own AI can run this check too, with no account, over Convertlyft’s MCP server. Connect your own AI tool to Convertlyft over MCP.

Run a free SEO audit of the same site → See all free SEO tools → What an SEO audit checks →