The new open standard from Google, Microsoft and Hugging Face decides which sites AI agents can discover. I check yours against it in seconds, free.
no signup · nothing stored · or paste a file / draft one from scratch
Answer in plain English. I type up a valid ai-catalog.json as you go. Everything happens in your browser.
It must be reachable at https://yourdomain.com/.well-known/ai-catalog.json. The folder is literally called .well-known, dot included.
.well-known/ai-catalog.json inside your published folder and redeploy..well-known folder in the site root (the same level as wp-content)..well-known path, or proxy it through Cloudflare./var/www/html/.well-known/ai-catalog.json.Then run the check at the top and I will confirm the job is done.
Already have a catalog file? Paste it here and I examine it against the official schema, entirely in your browser.
AI assistants increasingly act as agents: they book, buy, compare and research on a person's behalf. In 2026, Google, Microsoft and Hugging Face published Agentic Resource Discovery (ARD), a standard way for a website to declare what it offers in a form agents can read: one small file at /.well-known/ai-catalog.json.
Sites that publish the file can be discovered and used by agents. Sites that do not are, to an agent, a locked shop with no sign. I check the lock, and I write the sign.
The ARD specification is real and open, developed by Google, Microsoft and Hugging Face with collaborators including Cisco, GitHub, Nvidia and Salesforce. Agent Ardvark is an independent free tool built on the official spec and its published JSON Schema. Not affiliated with the ARDS project. No data is stored; domain checks pass through this site's server only to fetch the file, and pasted or generated files never leave your browser.