Early agent standards: what they are and why we don't grade them
New proposals no major AI agent has confirmed it uses yet. Not graded. The free check looks for 12 of them and shows what it finds, but they never count toward your fixes.
Why it matters
New files and headers for AI agents appear every few weeks. Some checkers give points for them; we don't. No major AI agent has publicly confirmed it uses any of these 12 to find or cite pages, so the free check shows what you have without grading it. We recheck each one weekly and will update this page if that changes.
| Standard | What it is | Status (Oct 2026) | Who reads it today |
|---|---|---|---|
| AI use preferences (Content Signals) | Lines in robots.txt that say whether AI may use your content for search, AI answers or training | Cloudflare policy, launched September 2025 | Cloudflare adds it to robots.txt files it manages. No AI company has publicly said it honours it. |
| AI use preferences (IETF draft) | The IETF's version of the same idea | IETF working-group draft | No crawler has committed to it |
| Content licence for AI (RSL) | A machine-readable licence for your content | RSL 1.0, published December 2025 | No AI crawler has publicly said it honours it |
| Markdown version for AI agents | Send Markdown when an agent asks for it | Cloudflare feature (beta, Pro and Business plans) | No AI search engine has confirmed using it |
| MCP server card | Describes your MCP server before an agent connects | MCP SEP-2127, Final | MCP clients and registries |
| Agent Skills index | Lists skills an agent can load | Cloudflare draft, version 0.2.0 | Agents that load skills. Few read the index. |
| API catalog | Lists your public APIs | RFC 9727 (IETF standard), June 2025 | API tools. Few consumers. |
| A2A agent card | Describes an AI agent others can call | A2A protocol, a Linux Foundation project | Agent platforms that call other agents |
| Agent checkout profile (UCP) | Lets AI agents check out on a shop | Universal Commerce Protocol, open spec | Partners listed on ucp.dev include Google and Shopify. No confirmed use for finding or citing pages. |
| Agent resource manifest (ARD) | Lists MCP servers, agents and skills for discovery | ARD v0.91, proposal (August 2026) | Discovery services. Adoption is early. |
| Tools AI agents can use (WebMCP) | Lets a page offer tools to an in-browser agent | W3C Community Group draft. Early preview in Chrome. | Chrome's preview and agents built on it |
| Personal agent sign-in (poppy.json) | Says how personal agents can work with your service | Draft 0.1 (October 9, 2026). Draft 0.2 has breaking changes. | Brand new. Adoption unknown. |
How to fix
Nothing here needs fixing. None of these is graded, and not having them costs you nothing today.
- Fix the graded checks first: AI agents allowed in, your firewall, content without JavaScript, Google's indexing rules and your buyers' words.
- Add one of these only if it fits what you already do:
- You run a public API: an API catalog.
- You run a remote MCP server (a way for AI agents to use your tools): an MCP server card.
- You offer an agent other agents can call: an A2A agent card.
- You run a shop and want Google's agent checkout: look at UCP.
- You want to state your AI use or licensing terms: Content Signals, the IETF Content-Usage line, or RSL. These state your wishes; they don't block anything. To block a bot, use robots.txt or your firewall.
- Serve a real JSON or text file at the path, not an HTML page. The free check counts an HTML page there as not found.
- Recheck when a new version ships. Follow the spec link to see what changed.
How to check it worked: run the free check again. In the "Early AI agent standards" group, the standard shows as found, for example "You publish an API catalog." The group stays Not graded.
What the free check looks for
- Read robots.txt for Content-Signal, Content-Usage, License and Agentmap lines, and the page headers for Content-Usage and an RSL Link header.
- Ask for the page with Accept: text/markdown and see if Markdown comes back.
- Fetch the well-known files: MCP server card (old and new paths), Agent Skills index, API catalog, A2A agent card, UCP profile (shop pages only), ARD manifest and poppy.json. An HTML page at those paths counts as not found.
- Scan the page for WebMCP tools (form toolname attributes and modelContext.registerTool).
- Show found, not found or not a valid file. Never pass or fail.
Caveats
- — These standards change fast. Paths and versions moved during 2026 (MCP server cards, ARD, poppy.json). We recheck them weekly.
- — Found does not mean any AI agent reads it. Not found costs you nothing today.
- — Several only matter for certain sites: an API, an MCP server, an agent others call, or a shop.
Studies and docs
2025
Standards Track RFC, published June 2025. Defines /.well-known/api-catalog.
2025
Status Recommendation, published 2025-12-10. Licences can be linked from a robots.txt License: line, an HTTP Link header or an HTML link.
2026
Working-group Internet-Draft (version 05, last updated 2026-08-18). Defines a Content-Usage HTTP header and robots.txt rule.
See what AI agents can find on your site
Our free AI visibility check shows whether AI agents can reach, read and quote your site, with each fix graded by evidence. There's no overall score, and no one can promise citations.
Free. No password.