What is an Agent Card in the A2A protocol?
An Agent Card is the JSON document an A2A server publishes to describe itself, including its identity, endpoint, supported features, authentication schemes and skills, so other agents can discover it and decide whether to delegate a task.
Checked
In the Agent2Agent protocol, one agent needs to know what another can do and how to reach it before handing over a task. The Agent Card answers both. The A2A documentation calls it a business card for an A2A server: a JSON file listing who the agent is, where its service lives, which protocol features it supports, how to authenticate and which skills it offers.
This page covers the card itself. The protocol, its governance and how it relates to MCP are covered on the A2A protocol page.
What a card contains
Identity
A name, description and provider for the agent.
Interfaces
Since v1.0, a supportedInterfaces list replaces the single url field. Each entry carries its own URL, protocol binding and protocol version, and the first entry is the primary endpoint.
Capabilities
Flags such as streaming, push notifications and whether an extended card is available. A client calling an operation the card does not declare gets an unsupported-operation error.
Security schemes
How callers authenticate, with options including API keys, HTTP auth, OAuth 2.0, OpenID Connect and mutual TLS.
Skills
Each skill has an id, name, description, input and output modes and examples, so a calling agent can judge fit before sending work.
How agents find cards
Well-known URI
Servers host the card at /.well-known/agent-card.json on their domain, following RFC 8615. Simple and automatic, but suited to open or domain-controlled settings.
Curated registries
A central catalogue that can be searched by skill and governed with access controls. The specification does not define a standard registry API.
Direct configuration
Hard-coded URLs or config files for fixed relationships. Simple, but every card change means reconfiguring the client.
Signing, private details and caching
Version 1.0 added a signatures array. Signatures use JWS (RFC 7515) over a canonical form of the card produced with JSON Canonicalization (RFC 8785), and clients should verify them when present.
Cards can leak information, such as internal endpoint URLs or descriptions of sensitive skills. The documentation recommends authenticated extended cards for the detailed version, protected endpoints, registries that show different cards to different callers, and dynamic credentials instead of secrets written into the card. Servers should send Cache-Control and ETag headers, and clients should use conditional requests rather than re-downloading.
What it is confused with
An MCP tool list
An MCP server lists tools a model calls directly. An Agent Card advertises an agent that takes whole tasks and decides how to do them.
Agent Skills
SKILL.md folders package instructions for one agent to load. A2A skills are entries on a card describing what a remote agent offers.
An agent registry
A registry stores and serves many cards. The card is the entry, the registry is the directory.
How Motiontic's engineers apply it
When a client's agent has to accept work from another team or company, Motiontic's engineers treat the Agent Card as a public contract: skills written as precisely as an API reference, the detailed card behind authentication, signatures turned on, and the card versioned so callers can cache it safely.
Common questions
Where is an A2A Agent Card published?
Usually at https://{domain}/.well-known/agent-card.json, following RFC 8615. Cards can also be served through a registry or configured directly in the client.
What changed in the Agent Card in A2A v1.0?
Version 1.0 replaced the top-level url with a supportedInterfaces list, moved the protocol version into each interface, moved the extended card flag into capabilities and added JWS signatures.
Can an Agent Card contain secrets?
It should not. The documentation recommends keeping sensitive details in an authenticated extended card and using dynamic credentials rather than static secrets in the card.
Related
A2A protocol
The open protocol for agents talking to other agents, now under the Linux Foundation.
MCP vs A2A
The protocol for tools and data, compared with the protocol for agents talking to agents.
Agent identity
Each agent gets its own credential, permissions and audit trail.
MCP server
The program that wraps a system so any MCP-compatible agent can use it.
AI agent development
Agents that do real work for users, with the evals, traces and guardrails to keep them in production.
Running agents that have to hold up?
Describe the agent, the systems it touches and what it must never get wrong. The first call is with a lab engineer, who will say plainly whether it is ready for production, and whether Motiontic is the right team.