AgentRoot
$ publish
Open · DNS-native · No gatekeepers
The DNS of the
agentic web

One domain. One manifest. Every agent capability — agents, MCP servers, skills, A2A endpoints — discoverable and verified via DNS.

$ ls /registry$ publish manifest
agentroot
$ npx agent-root search database --type mcp domain type name ────────────────────────────────────────────── acme.io/db-tools MCP DataTools sql.dev/query MCP SQL Query data.co/analytics MCP Analytics $ npx agent-root install acme.io/db-tools ✓ DNS verification passed ✓ MCP config generated for Claude Code ✓ 3 tools available: query, visualize, export $
Works with
Claude CodeCodex CLIGemini CLIGitHub CopilotCursor
// two_ways_in.md

Two ways in.

AgentRoot has two audiences. Find the path that's yours — and skip the rest.

// for_consumers
Find capabilities

You're wiring up an AI tool, building an agent, or running one — and you need an MCP server, a skill, an A2A endpoint, or a payment record that already exists somewhere. Browse the public registry by domain or search by capability, then install with one command.

Browse the registry →
// for_builders
Publish capabilities

You own a domain and you're shipping agents, MCP servers, skills, or A2A endpoints. Declare them in one agentroot.json manifest, add an _agentroot DNS TXT record, and the registry picks you up. No accounts, no API keys, no approval queue.

Read the publish guide →
// the_trust_problem

In the age of agents, provenance is everything.

AI agents are about to use tools on your behalf — calling APIs, moving money, accessing data. The question isn't "what can this tool do?" It's "who published it?"

When an AI agent finds a tool called "payments", how does it know it's legitimate? A GitHub username doesn't prove anything. An npm package name is just a string. But a domain? That's identity you can verify.

stripe.com has been Stripe's domain since 2010. It's in the internet's public directory. Nobody can fake it. Nobody can squat on it without being caught. Domain ownership is the strongest proof of identity on the web — and AgentRoot uses it as the foundation for trust.

One DNS record. That's your declaration of existence. We read it, verify it, and make your capabilities discoverable to every AI tool on the planet. No accounts. No gatekeepers. No approval process.

Which one do you trust?
✓stripe.com/payments
⚠random-guy.com/stripe-payments

Domain ownership is the identity layer the agentic web needs. AgentRoot makes it the foundation of agent discovery.

// install.sh

$ install — one command, any tool.

Find agents, MCP servers, and skills in the registry and install them with a single command. Works with Claude Code, Codex CLI, Gemini CLI, Cursor, and more.

$ npx agent-root install stripe.com/payments
# resolves via DNS, auto-detects your AI tool, configures it
# or install the CLI globally first
$ npm install -g @agent-root/cli
# Personal (all your projects) npx agent-root install stripe.com/payments --tool claude # Project-scoped (shared with team via git) npx agent-root install stripe.com/payments --tool claude --project
# Global npx agent-root install stripe.com/payments --tool codex # Project-scoped npx agent-root install stripe.com/payments --tool codex --project
# Project-scoped (cross-tool standard) npx agent-root install stripe.com/payments --tool gemini # Works alongside Codex CLI skills npx agent-root install stripe.com/payments --tool gemini --project
# Downloads to .cursor/skills/ npx agent-root install stripe.com/payments --tool cursor # Reference from your .cursorrules file
// publish.sh

$ publish your manifest in 3 steps.

Have agents, MCP servers, or skills? Declare them all in one manifest. No accounts. No approval. Just DNS.

01

Create a manifest

One JSON file listing your agents, MCP servers, skills, and A2A endpoints. Host it at .well-known/agentroot.json.

02

Add a TXT record

One DNS record: _agentroot.yourdomain.com pointing to your manifest. Takes 2 minutes.

03

Submit & done

AgentRoot verifies your DNS, fetches your manifest, indexes all records. Discoverable everywhere.

// why_dns.md

Built on the web's oldest infrastructure.

DNS has been running for 40 years. It's decentralized, public, and already trusted by every device on earth. We're using it — not replacing it.

Decentralized

No single company controls this

DNS is the internet's public directory. No one owns it. Your domain, your capabilities, your rules.

Verifiable

Domain = identity

If you control the domain, you control what's published under it. No impersonation possible.

Multi-type

Agents, MCPs, skills, A2A, payments

One manifest declares everything. Like DNS has A, MX, and TXT records, AgentRoot has agent, mcp, skill, a2a, and payment records.

Hierarchical

Domains & subdomains

Each domain has its own manifest. Subdomains publish their own records independently, just like DNS delegation.

$ man agentroot →

// roadmap.md

Where this is going.

Building the discovery layer for the agentic web. Openly.

live
Manifest model — agents, MCPs, skills, A2A in one manifest
One _agentroot TXT record. One manifest. All your agent capabilities — agents, MCP servers, skills, A2A endpoints — indexed and discoverable.
live
Search — find capabilities across the registry
Free-text search across every indexed manifest. Filter by record type (agent, MCP, skill, A2A, payment) from the CLI, MCP server, or web registry.
next
Continuous DNS monitoring
AgentRoot keeps your manifest fresh by re-checking DNS on a rolling schedule. Add or update an _agentroot record and the registry picks it up — no resubmission needed.
v3
Agentic tooling & integrations
First-class plugins for Claude, OpenAI, Cursor, Codex, Gemini, and other tools — so every agent runtime can resolve and install AgentRoot capabilities natively.
v4
Semantic search & analytics
Find capabilities by what they do, not just their name. "Find me an MCP server that queries databases." Plus per-record analytics — installs, resolutions, and usage trends.