Is your website agent-ready?
AI agents now visit your site the way search engines once did — except they cannot guess. Type a URL and we check the public surface agents actually look for: the files, headers, and conventions that make a site machine-negotiable. Nothing to install, nothing to sign up for.
We fetch like a plain, logged-out HTTP agent — no cookies, no JavaScript rendering, nothing authenticated.
Curious? — it takes its own test.
What we check
Eighteen checks in five groups. Four groups are scored, weighted by how much they decide whether an agent can discover, read, respect, and act on your site — the first two are table stakes every site should clear; the last one separates readable sites from usable ones. The fifth group, agentic commerce, detects product, feed, and checkout-protocol signals without scoring them: nothing found is the normal result for a site that does not sell online, so it never counts against you.
Discovery basics
Can anything find and make sense of the site at all: a robots.txt, a declared sitemap, real page metadata.
Content agents can use
Machine-usable versions of the content itself: markdown negotiation and an llms.txt / agents.md index.
Bot policy
Explicit rules for AI crawlers and content signals, so intent is stated rather than guessed.
Agentic surface
The emerging protocols agents use to act, not just read: MCP, A2A, skills, API catalogs, auth, bot signing.
Agentic commerce
Detection only, never scored: the product, feed, and checkout-protocol signals a shopping agent looks for. "Nothing found" is the normal result for a site that does not sell online.
What a score means — and what it does not
The check reads only what a logged-out agent would see. No cookies, no JavaScript rendering, no authenticated crawl, no content-quality judgement. A perfect score means your site is discoverable and machine-negotiable — it does not mean every agent will love your content, and it says nothing about pages hidden behind logins.
The scoring is deliberately simple: each scored check carries a weight, a pass earns it, a warning earns half, a failure earns nothing. Warnings are usually conventions skipped rather than walls hit — a sitemap that exists but is not declared, a title without a description. The agentic-commerce checks carry no weight at all: they report what they found (product markup, a declared feed, an ACP or UCP discovery file) and nothing more — detection, never a compliance verdict.
And the honesty clause: most of the checks in the last group are emerging protocols. Failing them is normal today. Publishing them is a statement that agents are welcome to do more than read — which is exactly the bet this company is built on.
This site takes its own test
Every check the tool runs is something thethirdwaycompany.com implements itself — the content signals in robots.txt, the llms.txt index, markdown negotiation on the homepage (ask for Accept: text/markdown and you get markdown), the MCP server card, the A2A agent card, agent skills, the RFC 9727 API catalog, and auth.md for how agents authenticate (spoiler: they mostly do not need to). The agentic-commerce group, meanwhile, honestly reports nothing found — we sell engagements, not catalog products, so there is no product markup or feed to detect. Run the check on thethirdwaycompany.com above and watch it pass its own exam.
Selling online and want the commerce-shaped side — product markup, feed declarations, ACP/UCP discovery — handled properly? Read the companion guide, Can an AI agent buy from you?, then decide whether it is work for us or for your team.
Want the same surface for your site without doing it by hand? That is a piece of work we do for clients — usually inside a week, scored by this same tool before and after. Or talk to us first; the check is free either way.
One email a month on practical AI adoption
What we ship for clients, what the tools can honestly do this month, and one thing worth trying in your own team — written by us, no hype, unsubscribe anytime.