Agent runtime

Agent Card

A structured description of what an AI agent can do, how to talk to it, and what trust it requires — the public face of an A2A-compatible agent.

Operating principle

Production AI is not a prompt. It is a system of context, tools, permissions, traces, evals, and feedback loops.

What it is

An Agent Card is metadata: a machine-readable description of an agent's name, what tasks it accepts, what schemas its inputs and outputs follow, what authentication is required, and what it costs to call. Think of it as the agent's business card — published by the agent, read by other agents (or by humans setting up integrations) to decide whether to use it.

Why it matters

If agents are going to collaborate at runtime instead of being wired together at design time, they need a way to advertise themselves. The Agent Card is the answer — a standard format that any A2A-compatible client can read.

What's typically in it

Identity (name, version, owner), capabilities (which task types the agent handles), schemas (request and response shapes), trust requirements (authentication methods, scopes, rate limits), pricing or quota info, and discovery hints (URLs, endpoints). The format is part of the A2A specification.

Related resources