Every check, and what we do about it.

The oldest and most widely honoured crawler convention, and the first place most well-behaved crawlers look. No file means no stated rules, and some crawlers treat that as a reason to stay away.

The plugin adds a robots.txt through WordPress, keeping whatever your SEO plugin already put there.

Still how a crawler finds every page rather than the ones it happens to link through.

WordPress publishes this already. We point at it from the Link header so an agent finds it without guessing the filename.

Everything below lives at a path an agent would have to guess. A Link header hands it over on the first request instead.

We send one RFC 8288 relation per document on every page, and write the same relations into the document head, so they survive a page cache or CDN that drops response headers. AI crawlers and readiness scanners are routed around the page cache and get the live headers.

Lets an agent find your documents from DNS, before it fetches anything. Two records do the work: an SVCB index at _index._agents and a TXT catalog pointer at _catalog._agents. Both need DNSSEC, because an unsigned answer is one an agent is told to distrust.

The plugin works out the exact records for your domain, checks whether they are live, publishes them for you when your DNS is on Cloudflare, and walks you through the one step it cannot do: the DS record at your registrar, with the screen location for the common ones.

A short, plain description of what the site is and which pages answer which question. A model reads it in one request instead of crawling forty pages.

Generated live from your pages and products, so it can never quote a price you stopped charging.

Your page as text, without the navigation, cookie banner and layout divs that make up most of the HTML a model has to pay for.

Same URL, same content, negotiated on the Accept header, with Vary set. Because a page cache replays stored HTML without reading Accept, the plugin routes AI crawlers and readiness scanners around the cache, and gives every page a markdown twin at ?format=markdown, advertised from the head, so the markdown is reachable through any cache or CDN.

A crawler that finds no group matching its own name has to guess. Naming them removes the guess.

We add a named group for eighteen AI crawlers, each with an explicit Allow.

States what AI systems may do with your work: appear in search, answer a question, train a model. Silence is read as consent by some and refusal by others.

We publish your three preferences inside every crawler group, so no agent has to infer them.

The same preference in the vocabulary the IETF working group is standardising. Cloudflare's Content-Signal draft expired in April 2026 without revision; AIPREF is on the Proposed Standard track and was revised in August. They are not interchangeable: one says ai-train=no, the other says train-ai=n.

We write both, in every crawler group, and send Content-Usage as a response header as well so it covers the files robots.txt cannot speak for. Emitting one and not the other means being unreadable to today's tooling or absent from tomorrow's standard.

Lets a verifier check that a signed request really came from you. This is the direction agent identity is going.

We generate an Ed25519 key on your own server and publish only the public half, with the correct RFC 8037 thumbprint.

The RFC 9727 index of every API this origin offers. Without it an agent has to discover your API by trying URLs.

Published as a linkset, listing the WordPress REST API, the WooCommerce Store API and the MCP endpoint.

Tells an agent what it can call and what comes back, in the format every tool already reads.

Generated from your live endpoints, including the product API when WooCommerce is active.

How an AI assistant discovers that your site has tools it can call, and which protocol versions they speak.

Published to the current MCP schema, with the endpoint, capabilities and supported versions filled in.

A card that describes a server nobody can connect to is worse than no card. This is the server itself.

We run a read-only MCP server on your site: site info, content search, live products and prices. No key, nothing writable.

Short instructions telling an agent how to do a specific job on your site, rather than leaving it to work one out.

We publish skills for finding a product and for understanding the site, each with a content digest so an agent can verify it.

The Agent2Agent card is how one piece of software finds out what another can be asked to do. Cloudflare's readiness model checks for it, and the AP2 payments profile is built on top of it.

We publish a card describing what this site can answer: site information, content search, and products when WooCommerce is active. It advertises nothing the site cannot actually do.

The ARD index: every machine-readable resource you offer, with the questions each one answers.

Built from what you actually publish, and announced in robots.txt as an Agentmap.

How software finds out whether it needs credentials and where it would get them.

RFC 9728 protected resource metadata, RFC 8414 authorization server metadata, and a readable auth.md alongside.

An agent can read your catalogue and still not buy anything. This says how a payment is made and where.

With WooCommerce active the plugin publishes ACP discovery, a UCP profile and an MPP offer over real WooCommerce orders, and x402 when you give it a wallet. Nothing is published until there is a real endpoint behind it, so a site without a store does not pretend to have one.