Can AI agents find you?

Agents are starting to discover capabilities at runtime instead of shipping with them hardcoded. Wellknown tells you whether yours are visible, whether what you publish is valid, and whether any registry actually returns you. Then it puts you in one.

The open registry

Every other ARD registry is closed. Hugging Face indexes only its own content, GitHub's Agent Finder is curated with no published submission route, Google's is not shipped. So you can serve a perfect catalog and have no door to knock on.

This one is open. Any domain can submit itself. No allowlist, no account, and no paid ranking, ever.

Submit any domain

curl -X POST https://wellknownhq.com/registry/submit \
  -H 'Content-Type: application/json' \
  -d '{"domain":"example.com"}'

Search it

curl -X POST https://wellknownhq.com/registry/search \
  -H 'Content-Type: application/json' \
  -d '{"query":{"text":"track a shipment"},"pageSize":5}'

Spec-shaped POST /search, with /explore for faceted browse and federation referrals to registries we do not operate. Also advertised at _search._agents.wellknownhq.com via DNS SRV.

Add it to your agent

Wellknown speaks MCP, so your agent can audit domains, validate catalogs, and search the registry directly. Stateless, protocol revision 2026-07-28.

claude mcp add --transport http wellknown https://wellknownhq.com/mcp

In claude_desktop_config.json:

{
  "mcpServers": {
    "wellknown": {
      "type": "http",
      "url": "https://wellknownhq.com/mcp"
    }
  }
}

In .cursor/mcp.json:

{
  "mcpServers": {
    "wellknown": {
      "url": "https://wellknownhq.com/mcp"
    }
  }
}

In .vscode/mcp.json:

{
  "servers": {
    "wellknown": {
      "type": "http",
      "url": "https://wellknownhq.com/mcp"
    }
  }
}

Any client that speaks remote MCP over HTTP. Most use one of the two shapes above; the endpoint is the only thing that varies:

https://wellknownhq.com/mcp

Transport streamable-http. No authentication. The server card is at /mcp/server-card and the discovery document at /.well-known/mcp.json, so a client that supports discovery can find it from the domain alone.

Plain JSON-RPC, no client required:

curl -X POST https://wellknownhq.com/mcp \
  -H 'Content-Type: application/json' \
  -H 'MCP-Protocol-Version: 2026-07-28' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call",
       "params":{"name":"audit_domain",
                 "arguments":{"domain":"notion.so"}}}'

Tool	What it does
`audit_domain`	Full visibility report for a domain, scored, with remediation steps
`validate_catalog`	Conformance findings for an `ai-catalog.json`, with JSON paths
`generate_catalog`	A conformant catalog, derived from what the domain already serves

Verification

Being in an index is not the same as being trustworthy, and the spec is explicit that a relevance score must never be read as a trust signal. So verification is a separate attribute you can filter on, never a ranking input. An unverified entry is never pushed down for being unverified.

Level	What it means	How
crawled	We found a public catalog and indexed it. We assert nothing beyond that it existed at that URL.	Automatic on submit
domain	The publisher proved control of the domain.	Free. One TXT record, see below
entity	A legal entity was checked, entries are signed, and revocation is published.	Paid

Prove your domain, free

curl -X POST https://wellknownhq.com/registry/verify/start \
  -H 'Content-Type: application/json' -d '{"domain":"example.com"}'

Add the TXT record it returns, then:

curl -X POST https://wellknownhq.com/registry/verify/check \
  -H 'Content-Type: application/json' -d '{"domain":"example.com"}'

Scoped to a _wellknown subdomain, so it never collides with SPF, DMARC, or the MCP Registry's own apex record.

What gets checked

Check	What it means
Discovery paths	All four the spec permits: `/.well-known/ai-catalog.json`, the `Agentmap:` directive in `robots.txt`, `<link rel="ai-catalog">`, and DNS records. Most tools check only the first.
Conformance	The official JSON Schema plus ten rules it cannot express: URN publisher matching your domain, `did:web:` identity, missing `representativeQueries`, duplicate identifiers, insecure URLs, unsigned trust manifests.
Registry indexing	Live queries using your own representative queries, reporting whether you come back and at what rank. Publishing is necessary. It is not sufficient.
Directory listings	The MCP directories that already carry discovery traffic today, so the answer is useful whether or not ARD adoption settles.

In our census of 45 major domains, including all eleven that co-authored the specification, exactly one served a valid catalog. Five served MCP's own mcp.json instead, in three mutually incompatible shapes. If you serve one of those, we read it and generate a conformant catalog from it, so there is nothing to author.