Documentation

The .agt agent namespace: register or migrate a name, set its records, publish a signed manifest, and resolve agents from anywhere. Start with the ramp that matches you.

I'm building an agent that uses .agt names

Resolve a name to its owner, endpoints and verified manifest, then connect to the agent behind it — from TypeScript, the CLI, an MCP client or raw JSON-RPC.

Quick start →Discover & connect to an agent →Examples →

I hold a .agt name

Migrate to Registry v2 for free, then publish a signed manifest and endpoint records so your agent is discoverable by name.

Migrating to v2 →Publish your agent's identity →Sign & verify manifests →

I'm building a client or tool

Add .agt lookups to Claude Code, Cursor, any MCP-compatible host, or your own runtime; or call the HTTP API and the contracts directly.

Add .agt to your client →Use with Claude Code →API Reference →

Start here

What .agt is, the problem it solves, the protocol layers, the self-sovereign AGT Registry, and naming conventions.

Resolve a name in a minute from five surfaces, publish your agent's identity, add .agt to your client.

Guides

Discover an agent, then connect to it

Find candidates in the directory, resolve and verify them, read the right endpoint, connect over MCP or A2A, and handle every verification outcome.

Publish your agent's identity

What a complete identity looks like, publishing from /manifest, writing records with viem, cast or ethers, hosting the manifest yourself, verifying, updating and rotating.

Sign and verify manifests

Canonical form, EIP-191 signing from a server or a wallet, three-way verification with real reasons output, CID checks, and the error states a client should handle.

Migrating to v2

For existing .agt holders: migrate your name to the AGT Registry — free and perpetual. The three migration paths, wallet requirements, the migration window, and FAQ.

Examples

Copy-paste snippets for the SDK, CLI, MCP server, HTTP API and on-chain reads and writes, with real mainnet output.

Reference

Records & Resolution

The on-chain record model (addr, text, manifest, endpoints, keys, wallet), how to set records from the site and from code, and how .agt names resolve.

The v3 manifest: a signed JSON document pointed to on-chain by the registry. Fields, vocabulary, signing and verification, and the resolution algorithm.

The @agtnames/resolver TypeScript package and the agt-resolve CLI. Chains, API, helpers, runtimes.

Use with Claude Code

The @agtnames/mcp server for any MCP-compatible client, with Claude Code as the worked example: install, tools, what verified means, connecting to an agent, troubleshooting.

Public HTTP routes exposed by agtnames.com — lookups, the directory, holder tools, registration and manifest pinning — with requests and responses.

Step-by-step walkthroughs: registering, migrating, resolving programmatically, agent-to-agent discovery, and exploring the directory.

System components and how the registry contracts, resolver, registration site, and resolver SDK fit together on Polygon.

Project

Where the AGT Registry stands today and what comes next — resolution surfaces, the agent layer, and network effects.