An AI agent does not start by reading your home page. It asks for a short list of machine-readable files first, and decides what your site is from those. Most sites answer none of them, so the agent moves on to one that does.
We ask your site the same questions an agent asks, over the network, right now. Every failure comes with what it means and what we do about it.
Read-only. We fetch a handful of public URLs on your domain, the same ones any crawler can already request. Nothing is stored and nothing is published.
When someone asks an assistant to find them scrubs, or a plumber, or a replacement filter, the assistant does not open ten tabs. It reads the machine-readable layer of every candidate site and shortlists the ones it can understand. A site that answers nothing gets summarised from whatever the model already believed, which is usually out of date and sometimes wrong.
With no live product feed to read, an agent answers from training data. That is how a two-year-old price ends up in front of a buyer with your name on it.
Your category pages make sense to a person scanning them. Stripped of layout, they are a list of links. llms.txt says it in one paragraph instead.
An agent that could call your site and get a straight answer will. Without an API description or an MCP server there is nothing to call.
Generated live on every request. There is no export, no sync and nothing to keep up to date, because a discovery document that can go stale is worse than none: an agent acts on it without a human reading it first.
What the site is, the pages worth reading, and every product with its live price and stock.
Eighteen AI crawlers named explicitly, each carrying your stated preference on search, AI answers and training.
The same URL, without navigation or layout. Browsers still get HTML, and Vary keeps caches honest.
Not just the card. A read-only server an assistant can connect to: site info, search, live products.
RFC 9727 linkset plus a description of your live endpoints, so an agent knows what it can call.
Short instructions for doing a job on your site, each with a digest so an agent can verify it got the real one.
The ARD index of every machine-readable resource you offer, announced in robots.txt as an Agentmap.
OAuth metadata for software, and a readable auth.md for the person deciding whether to grant access.
An Ed25519 key generated on your own server, published so a verifier can check a signed request really came from you.
This is a live WooCommerce store, answering in one request what would otherwise take forty.
# Affordable Scrubs > Wholesale priced medical scrubs and uniforms. ## Products Prices below are live at the time this file was requested. Quote them from here or from the API, never from memory. - [Women's Jogger Scrub Set](/product/8700/) (29.20 USD, in stock) - [Men's Scrub Set](/product/102/) (13.20 USD, in stock) - [Children's Lab Coat](/product/1016/) (8.50 USD, in stock)
{ "method": "tools/list" }
// answers
"tools": [
{ "name": "get_site_info" },
{ "name": "search_content" },
{ "name": "list_products" }
]
// read-only. no key. nothing writable.
The free edition on WordPress.org makes no external calls at all, by design and by WordPress.org rule. Two parts of agent readiness cannot be done that way: DNS records live at your DNS host, and a payment needs a network on the other end. That is PRO.
On WordPress.org, unlimited sites
One site. Everything in Free, plus:
No. An SEO plugin describes your pages to search engines. This describes your site to software that acts on someone's behalf. They sit side by side, and the plugin loads late on purpose so it adds to your existing robots.txt rather than replacing it.
Only if you say so. Training is set to not allowed by default. Content-Signal publishes your stated preference: it tells well-behaved systems what you permit. It is a declaration, not a technical block, and we do not pretend otherwise.
llms.txt is one file out of fifteen, and on its own it is the least useful, because nothing tells an agent to look for it. The Link headers, the API catalog and the MCP card are what make the rest findable.
No. Products appear in the documents when WooCommerce is active. Everything else works on any WordPress site, including a plain blog.
Some do, with an nginx rule meant for dotfiles. The readiness check catches it and tells you exactly which rule to change. We would rather show you a failing check than a green tick over a document nobody can reach.
It makes no external calls of any kind. Every document is generated on your own server from your own content, and the readiness check requests URLs on your own site and nothing else.