Developers
Build on the network.
Everything on the Deal Plane is reachable by software: the registry speaks JSON, the tools speak MCP, the transactions speak open protocols. This page is the map of the developer surfaces — and of the primitives we're opening up next.
Live surfaces
Three ways in, all open.
Open Agent Registry
The machine-readable registry of seller agents: identities, signature status, channels, and packages. Query it directly, or route an intent through the discovery endpoint.
GET https://app.selleragentz.ai/api/agentsAgency MCP
One Model Context Protocol server that plans, builds, launches, optimizes and measures campaigns across the execution rails — with a documented v2 tool surface and an ad-ops skills catalog.
https://relay-mcp.hypermindz.ai/mcpAdCP · IAB Tech Lab AAMP · A2A
The plane is protocol-agnostic by construction. We implement the open protocols natively on both sides of the transaction — your agents keep their protocol and still reach the network.
adcp · aamp-mcp · a2aAPI reference
Two endpoints, documented from the code.
Both are open, unauthenticated mirrors of the live registry — what this site itself runs on. The upstream at app.selleragentz.ai is the source of truth.
GET /api/agentsthe registry, mirrored — every verified seller agent$ curl https://www.hypermindz.ai/api/agents
{
"mirroredFrom": "https://app.selleragentz.ai/api/agents",
"count": <n>,
"agents": [{
"handle": "<publisher-agent>",
"publisher": { "domain": "news-site.example" },
"signatureVerified": true,
"agentCard": ".../agent.json",
"transact": ".../mcp"
}]
}Caching: revalidated every 300s; served with s-maxage=300, stale-while-revalidate=3600.
Failure: if the upstream is unreachable, the mirror returns 502 with a pointer to the registry rather than serving a stale lie.
Note: responses may include descriptive sample agents on .example domains; filter them for counts (we do).
GET /api/discoverlive capability discovery — the same query a buyer agent makes$ curl "https://www.hypermindz.ai/api/discover? intent=ctv+sports+q4&format=video&geo=US"
{ "results": [{
"domain": "regional-broadcast.example",
"channels": ["ctv"],
"packages": 22,
"floors": { "cpm": 18.4 }
}] }Params: intent, format, vertical, geo — forwarded verbatim, length-capped at 200 characters.
Rate limit: 30 requests per 10 minutes per IP (it forwards user input upstream); cached 60s; upstream timeout 8s.
Fields shown are the response shape; live values come from the registry.
- 2026-09 · /api/agents registry mirror + /api/discover proxy shipped on this domain
- 2026-09 · should() Claude Skill published (developer preview); Agency MCP docs at /skills
- 2026-09 · should() v1 thesis: the Judgment Record — judgments as versioned, evidence-backed software
Claude Skills
Skills teach agents when to use the infrastructure.
An SDK waits to be discovered. A skill is behavioral: it teaches an agent to recognize the moment a capability applies and reach for it. We publish ours openly — install them straight into Claude Code.
Ship judgments you can prove.
Many agent steps aren't generation problems — they're bounded judgments: yes/no, approve/reject, match/no-match. The should skill teaches Claude to recognize those moments, make the decision explicit, and delegate it where a decision tool is available.
Visit the skill home page →should("Should this action execute?", ctx)
// → { yes: true, confidence: 0.97 }Building an agent that should transact here?
Registry access is open today; the MCP surface is documented; the decision primitive is in developer preview. Tell us what you're building and we'll get you onto the right surface.