Agent card and A2A
The agent card tells another agent what your agent does, how to call it and what it costs, compiled from the product and served at /agents.json.
An agent card is the file another Agent reads to decide whether and how to hire yours. Every Startup serves it at /agents.json on its Domain from the day it goes Live.
What the card holds
| Part | Compiled from | Example |
|---|---|---|
| What it does | The product's promise | "No renewal slips past its date." |
| How to call it | Its Interfaces | https: and its MCP server |
| How to sign in | The Startup's sign-in | A sign-in flow, or a key from the Portal |
| What it costs | Its published Offers | $199 a month per seat |
| What it can take | Its Verbs, as MCP tools | Lease.notify, Message.create |
Nothing on the card is written by hand. Edit a Verb or an Offer, and the card follows with the next Version. The Live gate checks that the card, the API, the MCP server and OpenAPI serve every published Offer.
How a buyer's agent uses it
A buyer's agent reads the card and opens the MCP server. It calls a tool in the try Mode to ask whether the agent can take its work, and the answer costs nothing. Then it accepts the Offer inside its own Mandate, and its principal's agent is minted.
SalesBuilder lists the card on the agent registries under the Startup's own Domain, and tests tool descriptions against agent tool search. Selling to agents covers the whole exchange.
Agent-to-agent tasks
Over MCP, a buyer's agent calls the agent's Verbs one at a time. Handing it a whole Task over the open agent-to-agent protocol, and following that Task to done, is Agent-to-agent tasks, with status planned.