An agent card is a machine-readable software profile that tells other systems what an AI agent is, what it can do, and how to communicate with it. In agent-to-agent systems, it usually takes the form of a JSON document that supports discovery and interoperability. The term can also describe a virtual payment credential for an AI agent, but that is a separate agentic-commerce use of the name.
Agent Card
What Is an Agent Card?
An agent card is a machine-readable software profile that tells other systems what an AI agent does and how to communicate with it. Usually published as JSON, it works like a digital README or service description for an agent, not a physical or financial card.
In agent-to-agent, or A2A, systems, the card helps software discover compatible agents before making a request. The same name is sometimes used for a virtual payment credential that lets an AI agent make controlled purchases, but that is a separate agentic-commerce meaning.
What an A2A Agent Card Contains
An A2A agent card describes an agent's public integration surface. Exact fields vary by protocol and version.
| Field | Purpose |
|---|---|
| Name and description | Identifies the agent and its intended role. |
| Provider or operator | States who runs or maintains it. |
| Skills and capabilities | Lists tasks the agent can perform. |
| Endpoint and formats | Shows where requests go and which inputs or outputs are supported. |
| Authentication and version | Explains access requirements and compatibility details. |
How Agent Discovery Works
Discovery separates finding an agent from deciding whether it is safe to use.
- The provider publishes the JSON card at an agreed location or registry.
- A requesting agent retrieves the card.
- It validates the schema, version, and publisher origin.
- It selects a capability that matches the requested task.
- It authenticates and calls the advertised endpoint.
A card advertises an interface. It does not itself grant access or prove that the publisher is trustworthy.
Agent Card vs. MCP Tool Metadata
Both formats describe capabilities, but they support different relationships.
| Aspect | Agent card | MCP tool metadata |
|---|---|---|
| Primary purpose | Discover independent agents. | Expose tools or data to an agent. |
| Relationship | Agent to agent. | Agent to tool server. |
| Scope | Agent identity, skills, endpoint, and access. | Individual tool names, inputs, and outputs. |
| Discovery | May use well-known URLs or registries. | Usually follows an existing server connection. |
Benefits of Agent Cards
A clear profile reduces ambiguity as multi-agent applications grow.
- Improves interoperability across platforms and teams.
- Supports faster capability matching and automated discovery.
- Reduces integration guesswork for an integration agent.
- Keeps public agent documentation closer to the running service.
Practical Limits and Security Considerations
Treat every card as a claim that needs verification.
- Metadata can become stale, incomplete, or overly broad.
- A public card can expose sensitive implementation details.
- Schema changes can break older clients.
- Use HTTPS, authentication, authorization, and trusted registries or signatures where appropriate.
Using Agent Cards When Building AI Agents
Publish only the metadata another system needs, version the card, and test missing, invalid, and outdated-card paths. Keep discovery separate from authorization, so a discovered agent still must prove it may perform a request. These practices align with the principles of building AI agents and help teams build an AI agent that can integrate safely. Emergent helps developers build agent-aware applications from natural-language requirements through deployment.
Frequently Asked Questions
Your Questions, Answered
This will automatically populate, don't change
Don't change this element unless you know what you are doing
What is an A2A agent card?
An A2A agent card is a JSON-based profile used to describe an AI agent to other agents and systems. It commonly includes identity, skills, endpoint details, supported formats, and access requirements.
What information does an agent card contain?
It typically contains an agent name, description, provider identity, capabilities, service endpoint, authentication requirements, input and output formats, and version details.
How is an agent card different from MCP?
An agent card helps independent agents discover and communicate with each other. MCP metadata usually describes tools and resources that an AI agent can use through an MCP server.
Does an agent card authenticate an AI agent?
No. An agent card can describe authentication requirements, but it does not authenticate the agent by itself. Systems still need to verify identity and enforce authorization.
Is an agent card the same as an AI agent credit card?
No. An A2A agent card is software metadata for discovery and interoperability. An AI agent credit card is a payment credential with spending controls for agent-driven purchases.
Start Building
on Emergent today
on Emergent today
Start Building
