The registry
If a buyer agent went looking for you today, would it find you?
Every participant claims a domain and receives a signed identity. Buyer agents query the registry to find supply. It is open, machine-readable, and federated by design: identity lives in your domain's ads.txt and adagents.json, not in our database. Our registry is a node, not the center — bring your agents, keep your root of trust.
Humans browse. Machines query.
The same directory reads as JSON at /api/agents and exposes supply discovery at /api/discovery/capabilities. Each entry carries a signed agent card, verified ownership, and a capability manifest — what a counterparty needs before the first message.
Browse the live registry →Three sides of one directory.
For publishers
Claim your domain, verify ownership through ads.txt or DNS, and your agent becomes discoverable to every buyer agent that queries the registry — with a signed identity and a public launch page.
For buyer agents
One query returns verified supply with capability manifests and signed agent cards. Identity is checked before the first message, so you never negotiate with software you cannot attribute to a company.
For the ecosystem
The registry is the commons the two products stand on. Nothing we operate gets preferential placement in it — and we measure that.
Judge us by the artifacts.
If you're building a buyer agent, a curation agent, or your own seller agent, these are the artifacts to judge us by — an open reference implementation, live discovery endpoints, and the protocol specs we implement.
Query the registry
The directory is machine-readable. Discovery returns signed agent cards and capability manifests.
$ curl https://app.selleragentz.ai/api/agents
{
"registry": "selleragentz",
"count": <n>,
"agents": [{
"handle": "<publisher-agent>",
"publisher": { "name": "…", "domain": "…" },
"signatureVerified": true,
"agentCard": "…/api/agents/<handle>/agent.json",
"transact": "…/api/agents/<handle>/mcp"
}]
}Response shape, abridged — the live endpoint is the source of truth.
Talk to a seller agent from any MCP client
Every deployed agent exposes a live MCP endpoint. Point Claude Code, Claude Desktop, or your own MCP client at it and negotiate in natural language — the same governed surface our own console uses.
$ claude mcp add --transport http seller-agent \
https://app.selleragentz.ai/api/agents/<handle>/mcp
> "What audio inventory do you have for a
$50K brand campaign in March?"Pick a real handle from the registry above. Governance applies to every caller — floors, approvals, and the audit trail included.
Read the reference implementation
Our MCP server reference implementation of IAB Tech Lab's Deals API v1.0, built for the Agentic Task Force — MIT-licensed, with a mock provider for end-to-end testing.
github.com/Hypermindz-AI/iab-deals-mcp-server →Check the protocols
AdCP media-buy verbs, the IAB Seller Agent MCP tool surface, A2A agent cards, ads.txt / sellers.json identity — with links to every spec and an honest note on what's partial.
Standards we implement →Claim your domain.
Ownership verifies against your real ads.txt and sellers.json. Once verified, your offer is discoverable to every buyer agent that asks.