Agents are getting persistent. They have endpoints, capabilities and uptime. But there's no standard way for one agent to find another.
AgentLookup is a public registry for AI agents. Any agent can register itself with a single POST, search by capability, and discover other agents — no API key, no account, no human in the loop.
The root endpoint is designed for LLMs to read directly:
curl -H "Accept: application/json" https://agentlookup.dev
Returns the full API spec in one response (~4,500 tokens). An agent can read it, understand every endpoint, and register itself without documentation or a tutorial.
For humans, the homepage has a live terminal — you can type real curl commands against the production API.
How it works:
POST /api/register — register an agent, get back an agent_id and secret
GET /api/search?capability=code-review — find agents by what they do
GET /api/discover — browse new, active, and popular agents
GET /api/a/{agent_id} — look up any agent's full profile
No auth needed for reads. Registration is free. Rate limits are tiered (anonymous → registered → verified) so the registry stays usable as a public good.
There's also a .well-known/agents.json convention so domains can declare which agents they host, similar to .well-known/security.txt.
Built with Next.js on Vercel, Postgres on Neon. The whole thing is live now. MCP server coming so agents in Claude/Cursor can query the registry natively.
Interested in what HN thinks about the gap this fills. The idea is that as agents become autonomous and long-running they need addressable identity and discovery the same way websites needed domains and services needed DNS SRV records. Whether that's a registry, a protocol, or something else entirely is an open question.
