Creativity · Agent Protocol

A2A Agent Card — Capability Manifest Spec

In Google's A2A protocol, the Agent Card is the capability manifest — a JSON document, typically served at a well-known URL, that describes an agent to the outside world. It lists the agent's name, skills, endpoints, auth requirements, supported transports, and any streaming or long-running capabilities. Agent Cards make A2A discovery 'lazy': you don't register in a central directory, you just expose your card.

Protocol facts

Sponsor
Google (+ industry partners)
Status
proposed
Spec
https://google.github.io/A2A/
Interop with
A2A, JSON Schema, HTTP, OAuth

Frequently asked questions

What fields does an Agent Card include?

Typical fields are name, description, version, skills (each with input/output schemas), endpoint URL, supported protocols, authentication method, and capability flags like streaming or long-running task support.

How is an Agent Card discovered?

Usually by convention — fetched from a well-known path on the agent's host, or referenced from a registry. There's no central directory, which keeps A2A decentralised.

Is the Agent Card like an OpenAPI spec?

Philosophically similar — both describe callable capabilities — but Agent Cards focus on agent-shaped concepts (skills, streaming, long-running tasks) rather than generic REST resources.

Sources

  1. Google — A2A overview — accessed 2026-04-20