Which ones only read your robots.txt, which ones check llms.txt and MCP, and which one actually sends an agent to try to buy something.
This year, "is my website AI agent ready?" turned into a real question with real tools. Cloudflare launched a scanner, Google added an Agentic Browsing category to Lighthouse, and a handful of companies shipped their own free AI agent readiness checker right after. So I ran the same website, gifsy.fun, through every free checker I could find. The scores for that one site ranged from 20 to 100.
The short version: most of these tools check whether your site publishes the right signals. Only one checks whether an agent can finish a task. Both matter, but they answer different questions.
Here's what's actually worth your time:
- A free scanner from Cloudflare that checks the newest agent standards, including MCP and WebMCP
- A Lighthouse category already built into Chrome that you can run without leaving DevTools
- A no-signup checker from Frase that's the clearest about content and llms.txt
- A developer-first checker from Vercel that also runs as a CLI and hands you copy-paste fixes
- A tool that sends a real agent to do a real task on your site
Let's get into it:
1. Actuity
A real Actuity scan of gifsy.fun, start to finish: 87/100, 25 of 31 checks passed across 11 pages
Actuity sends a real AI agent to your website and asks it to do something a customer would do: find the price of a plan, add a product to the cart, send the contact form. Every other tool here is a checklist. A checklist can tell you your llms.txt exists. It can't tell you the agent clicked "Pricing", landed on a page where the numbers only load after a cookie banner, and quietly left.
It works in two parts. The free scan reads a sample of your public pages the way an agent would. On gifsy.fun it checked 11 of about 68 pages and gave it 87/100, with 25 of 31 checks passed. Then the agent analysis: you pick a task like "find the price of the main plan" and watch a real agent try it in a browser, step by step.
Free scan for any public URL, no login. Your first agent analysis is free.
Try it here: Actuity
2. Cloudflare's Agent Readiness Score (isitagentready.com)
Cloudflare's scanner gave gifsy.fun 20/100, Level 1 "Basic Web Presence"
This is the most standards-heavy checker on the list. It requests your homepage plus a few well-known paths and checks discovery (robots.txt, sitemap, Link headers), markdown versions of your pages, AI bot rules, and newer agent protocols like MCP server cards and WebMCP. Every check shows the raw response it saw, which makes debugging easy.
On gifsy.fun it was the harshest by far: 20/100. robots.txt and the sitemap passed, but there were no Link headers, no markdown version, no AI-specific robots rules, and 0 of 8 on API, auth, MCP and skill discovery.
The catch: most sites fail most of the protocol checks because the protocols are only a few months old. A low score here mostly means "you're early", not "agents can't use your site".
Free, no account needed.
Try it here: isitagentready.com
3. Lighthouse Agentic Browsing (in Chrome DevTools)
Lighthouse 13.5's Agentic Browsing category on gifsy.fun: 3 of 3 applicable audits passed
If you already run Lighthouse, this is the zero-effort option. Open DevTools, tick "Agentic Browsing" and analyze. It audits your accessibility tree, layout stability, your llms.txt and any WebMCP tools. The accessibility part is quietly the most useful check in this post: browser agents lean on the same structure screen readers do, so an unlabeled button breaks both.
gifsy.fun passed 3 of 3 applicable audits: a clean accessibility tree, a layout shift of 0 and a valid llms.txt. The four WebMCP audits were marked not applicable.
The catch: there's no 0 to 100 score, just a ratio of passed checks, and "not applicable" really means "you haven't adopted this yet". A perfect 3/3 says less than it looks.
Free, built into Chrome.
Try it here: Lighthouse Agentic Browsing docs
4. Frase Agent Readiness
Frase scored the same site 81/100, "Strong", 5 of 8 checks passed
This is the most readable checker if you care about content. It scores one URL on access (does real content show up without JavaScript, or does a bot challenge block it?), discovery (AI crawler policy, sitemap, llms.txt) and parseability (structured data, a markdown version). The "content without JavaScript" check alone is worth running: plenty of modern sites are a blank page to crawlers until the JS bundle loads.
gifsy.fun got 81/100. Server-rendered content, sitemap, llms.txt and structured data passed. It got a partial for not stating a position on AI crawlers in robots.txt, and failed both markdown checks.
The catch: it's built by an SEO company, so it leans toward "will AI cite my content" more than "can an agent complete a task".
Free, no sign-up.
Try it here: Frase Agent Readiness
5. Vercel's Is Agentic (is-agentic.com)
Vercel's Is Agentic gave gifsy.fun 75/100, "Ready with a few material gaps"
This is the one I'd hand to a developer. It's the only checker here that also runs from your terminal (npx is-agentic gifsy.fun). It guesses your site type (content, business, app or commerce), adjusts which checks count, and gives every gap a "Copy fix" prompt you can paste straight into your coding agent, with the curl command to verify it.
For gifsy.fun it inferred "App" and gave it 75/100: 4 of 7 essential checks and 6 of 12 recommended ones. The main miss was the 404 page, which returns the right status but no markdown body when an agent asks for one.
The catch: it's the strictest about markdown and APIs. A markdown 404 page and a published OpenAPI spec are a lot to ask from a small marketing site.
Free, no account, plus a free CLI.
Try it here: Is Agentic
What an AI agent readiness checker can and can't tell you
Here's how the same site did on each tool:
| Tool | What it tests | Result on gifsy.fun |
|---|---|---|
| Actuity | Free scan, then a real agent doing a real task | 87/100 (25 of 31 checks) |
| Cloudflare | Discovery, markdown, bot rules, agent protocols | 20/100, Level 1 |
| Lighthouse Agentic Browsing | Accessibility tree, CLS, llms.txt, WebMCP | 3/3 passed, 4 not applicable |
| Frase | Access, discovery, parseability | 81/100, 5 of 8 checks |
| Vercel Is Agentic | Essential, recommended and bonus checks by site type | 75/100 |
The same site got a 20 and a perfect 3/3 on the same afternoon. No tool is wrong: Cloudflare grades you on protocols that are months old, while Lighthouse and Frase grade the basics that already matter. Three of the five flagged the same gap on gifsy.fun, a missing markdown version of the page. What none of the scanners can see is the stuff that breaks in practice, like a pricing table hidden behind a modal or a form that fails silently. You only find those by watching an agent try.
Which one should you use first?
Pick by the kind of site you run:
- Blog or marketing site: Lighthouse first, then Frase. Fix the accessibility tree and make sure your content loads without JavaScript. Skip the protocol checks for now.
- SaaS with pricing and signup: Actuity first, with a task like "find the price of the main plan", then Is Agentic to turn the gaps into fixes your coding agent can ship.
- Online store: Actuity first with an "add to cart" task, then Cloudflare to see which agent commerce protocols you're missing.
- Docs, APIs or developer tools: Is Agentic first (OpenAPI, markdown, the CLI fits your CI), then Cloudflare for MCP and API discovery.
- Only have ten minutes: run Lighthouse, because it's already in your browser and the fixes help human visitors too.
Thanks for reading. If you've found another agent readiness tool that tests something these don't, tell me, I'd genuinely like to try it.
Originally published on Medium.
Top comments (0)