$ cat protocol.md

Declare your domain's agent capabilities in DNS. One TXT record. One JSON file. That's it.

// how_it_works

Three layers

The AgentRoot protocol has three layers. DNS proves ownership. The manifest declares capabilities. The registry indexes everything for search and discovery.

Layer 1: DNS

_agentroot.example.com TXT "v=ar1 manifest=https://..."

→

Layer 2: Manifest

{ "domain": "example.com", "records": [ { "type": "agent", ... }, { "type": "mcp", ... } ] }

Layer 3: Registry (optional)

// Search, indexing, CLI, MCP server, A2A // Convenience layer — protocol works without it

Step 1: Add a TXT record at _agentroot.<your-domain>. It points to a JSON file.
Step 2: The JSON file lists your agent capabilities — agents, MCP servers, skills, A2A endpoints.
Step 3: AgentRoot verifies the DNS record and indexes your manifest.

// dns_record

The DNS record

Add a TXT record at _agentroot.<your-domain>:

# Manifest URL mode — points to a JSON manifest
_agentroot.example.com  TXT  "v=ar1 manifest=https://example.com/.well-known/agentroot.json"

Two fields. v=ar1 is the protocol version. manifest=<url> points to your manifest. For single-record setups, you can skip the manifest entirely with inline mode.

# Inline mode — single record, no manifest needed
_agentroot.example.com  TXT  "v=ar1 type=agent name=MyBot endpoint=https://example.com/api"

Full DNS record docs →

// manifest

The manifest

Host a JSON file at https://<domain>/.well-known/agentroot.json. It declares everything your domain offers.

{
  "domain": "example.com",
  "records": [
    {
      "type": "agent",
      "id": "assistant",
      "name": "My Assistant",
      "description": "A research assistant",
      "endpoint": "https://example.com/agent"
    },
    {
      "type": "mcp",
      "id": "db-tools",
      "name": "DataTools",
      "description": "Database query and visualization tools",
      "endpoint": "https://example.com/mcp",
      "transport": "sse"
    }
  ]
}

Full manifest docs →

// record_types

Record types

Five built-in types. The protocol schema is a closed enum at the validator layer — records with any other type are rejected.

Type	What It Is	Key Fields
agent	An AI agent that accepts tasks and returns results	`endpoint`, `protocol`, `capabilities`
mcp	A Model Context Protocol server exposing tools	`endpoint`, `transport`, `tools`
skill	A collection of SKILL.md instruction files	`index` or `skill_md`
a2a	An Agent-to-Agent protocol endpoint	`endpoint`, `capabilities`
payment	A payment endpoint advertising agent-payment protocols and accepted assets	`endpoint`, `protocols`, `methods`, `assets`