A2A Agent Card

Before writing code

Fetch live docs:

1. Fetch https://a2a-protocol.org/latest/specification/ and look for the Agent Card schema section
2. Web-search site:github.com a2aproject A2A agent card schema for the JSON Schema definition
3. Web-search site:github.com a2aproject a2a-samples agent-card for real-world Agent Card examples
4. Fetch SDK docs for Agent Card builder/helper classes

Conceptual Architecture

What the Agent Card Is

The Agent Card is a JSON document that describes an A2A agent to potential clients. It's the discovery and capability advertisement mechanism — analogous to an OpenAPI spec for REST APIs, but for agent-to-agent communication.

Hosting

By convention, Agent Cards are served at:

{agent_base_url}/.well-known/agent-card.json

Agents can also be discovered via:

• Agent registries (centralized directories)
• Direct configuration (hard-coded URLs)
• Referral from other agents

Key Sections

Identity

• name — Human-readable name
• description — What the agent does
• url — The agent's A2A endpoint (where JSON-RPC requests go)
• version — Semantic version of the agent
• protocolVersion — The A2A protocol version the agent supports (required)
• provider — Optional organization/owner information

Capabilities

Declares what protocol features the agent supports:

• streaming — Can the agent handle message/stream?
• pushNotifications — Can the agent send push notifications?
• stateTransitionHistory — Does the agent maintain full task state history?

Skills

Array of skills (capabilities) the agent offers:

• id — Unique skill identifier
• name — Human-readable skill name
• description — What the skill does
• tags — Categorization tags
• examples — Example inputs/prompts for this skill

Skills help client agents understand what tasks they can delegate.

Authentication

Declares required authentication via two top-level fields:

• securitySchemes — A map/object of named security scheme definitions (e.g., apiKey, http with scheme: bearer, oauth2, openIdConnect)
• security — An array of required scheme references (e.g., [{ "schemeName": [] }])
• Each scheme has type-specific configuration (header name, OAuth URLs, OIDC discovery URL, etc.)

Input/Output Modes

• defaultInputModes — MIME types the agent accepts (e.g., text/plain, application/json, image/png)
• defaultOutputModes — MIME types the agent can produce

Agent Card Design Principles

• Be specific in descriptions — Other agents (LLMs) will read this to decide whether to delegate tasks
• Skills should be granular — Each skill represents one capability, not the entire agent
• Examples are critical — They help client agents understand how to formulate requests
• Declare all required auth — Clients need to know upfront what credentials they need
• Version meaningfully — Update version when skills or capabilities change

Use Cases

• Public-facing agent discovery (web-crawlable Agent Cards)
• Internal agent registry for enterprise multi-agent systems
• Marketplace listings where agents advertise their skills
• Framework integration points (ADK, LangGraph, CrewAI read Agent Cards)

Best Practices

• Validate the Agent Card against the official JSON Schema
• Serve with appropriate CORS headers if agents run in browsers
• Include meaningful tags for skill categorization and search
• Keep descriptions concise but informative — they're read by LLMs
• Test that /.well-known/agent-card.json is accessible from client agents
• Update the Agent Card when adding/removing skills or changing capabilities

Fetch the specification for the exact Agent Card JSON Schema, all field types, and required vs optional fields before implementing.