---

name: design-a2a-agent-card locale: caveman source_locale: en source_commit: 82c77053 fence_basis_commit: 82c77053 translator: "Julius Brussee homage — caveman" translation_date: "2026-04-24" description: > Design an A2A Agent Card (.well-known/agent.json) manifest describing agent capabilities, skills, authentication requirements, and supported content types. Use when building an agent that must be discoverable by other A2A-compliant agents, exposing capabilities for multi-agent orchestration, migrating an existing agent to the A2A protocol, defining the public contract for an agent before implementation, or integrating with agent registries that consume Agent Cards. license: MIT allowed-tools: Read Write Edit Grep Glob metadata: author: Philipp Thoss version: "1.0" domain: a2a-protocol complexity: intermediate language: multi tags: a2a, agent-card, manifest, capabilities, interoperability

Design A2A Agent Card

Make A2A Agent Card. Advertises agent identity, skills, auth, capabilities. Other agents find it.

When Use

• Build agent others must discover via A2A
• Expose agent capabilities for multi-agent orchestration
• Migrate existing agent to A2A (Agent-to-Agent) protocol
• Define public contract before implementation
• Integrate with agent registries that read Agent Cards

Inputs

• Required: Agent name + description
• Required: Skill list (name, description, input/output schemas)
• Required: Base URL where agent hosted
• Optional: Auth method (none, oauth2, oidc, api-key)
• Optional: Content types beyond text/plain (e.g., image/png, application/json)
• Optional: Capability flags (streaming, push notifications, state history)
• Optional: Provider org name + URL

Steps

Step 1: Set Agent Identity + Description

1.1. Pick identity fields:

{
  "name": "data-analysis-agent",
  "description": "Performs statistical analysis, data visualization, and report generation on tabular datasets.",
  "url": "https://agent.example.com",
  "provider": {
    "organization": "Example Corp",
    "url": "https://example.com"
  },
  "version": "1.0.0"
}

1.2. Write clear, actionable description. Answer:

• What domains agent covers?
• What tasks handles?
• What limits?

1.3. Set canonical URL where Agent Card served at /.well-known/agent.json.

Got: Full identity block: name, description, URL, provider, version.

If fail: Agent covers many domains? Decide: one agent, many skills? Or many agents, focused scope? A2A prefers focused agents, clear boundaries.

Step 2: List Skills with I/O Schemas

2.1. Define each skill:

{
  "skills": [
    {
      "id": "analyze-dataset",
      "name": "Analyze Dataset",
      "description": "Run descriptive statistics, correlation analysis, or hypothesis tests on a CSV dataset.",
      "tags": ["statistics", "data-analysis", "csv"],
      "examples": [
        "Analyze the correlation between columns A and B in my dataset",
        "Run a t-test comparing group 1 and group 2"
      ],
      "inputModes": ["text/plain", "application/json"],
      "outputModes": ["text/plain", "application/json", "image/png"]
    },
    {
      "id": "generate-chart",
      "name": "Generate Chart",
      "description": "Create bar, line, scatter, or histogram charts from tabular data.",
      "tags": ["visualization", "charts"],
      "examples": [
        "Create a scatter plot of height vs weight",
        "Generate a histogram of the age column"
      ],
      "inputModes": ["text/plain", "application/json"],
      "outputModes": ["image/png", "image/svg+xml"]
    }
  ]
}

2.2. Each skill needs:

• id: Unique ID (kebab-case)
• name: Human-readable name
• description: What skill does, 1-2 sentences
• tags: Keywords for discovery
• examples: Natural-language task examples that trigger skill
• inputModes: MIME types skill takes
• outputModes: MIME types skill produces

2.3. Skill boundaries must be clear, no overlap. Each task → one skill.

Got: Skills array. Each entry: id, name, description, tags, examples, I/O modes.

If fail: Skills overlap big? Merge into broader skill with more examples. Skill too broad? Split into focused sub-skills.

Step 3: Config Auth

3.1. Pick auth scheme by deploy context:

No auth (local/trusted network):

{
  "authentication": {
    "schemes": []
  }
}

OAuth 2.0 (best for prod):

{
  "authentication": {
    "schemes": ["oauth2"],
    "credentials": {
      "oauth2": {
        "authorizationUrl": "https://auth.example.com/authorize",
        "tokenUrl": "https://auth.example.com/token",
        "scopes": {
          "agent:invoke": "Invoke agent skills",
          "agent:read": "Read task status"
        }
      }
    }
  }
}

API Key (simple shared secret):

{
  "authentication": {
    "schemes": ["apiKey"],
    "credentials": {
      "apiKey": {
        "headerName": "X-API-Key"
      }
    }
  }
}

3.2. Pick minimum viable auth for env:

• Local dev: none
• Internal service: apiKey
• Public-facing: oauth2 or oidc

3.3. Document token/key provisioning in provider section or external docs.

Got: Auth block matches deploy security needs.

If fail: No OAuth 2.0 infra? Start with API key, plan migration. Never deploy public agent with none auth.

Step 4: Declare Capabilities

4.1. Declare protocol features agent supports:

{
  "capabilities": {
    "streaming": true,
    "pushNotifications": false,
    "stateTransitionHistory": true
  }
}

4.2. Set each flag by impl readiness:

• streaming: true if agent supports SSE streaming via tasks/sendSubscribe. Real-time progress for long tasks.
• pushNotifications: true if agent can send webhook callbacks on state change. Agent stores + calls webhook URLs.
• stateTransitionHistory: true if agent keeps full state transition history (submitted, working, completed). Good for audit.

4.3. Only set true if impl fully supports. Fake flags break interop.

Got: Capabilities object. Flags match real impl.

If fail: Unsure if capability coming? Set false. Add later. Removing capability = breaking change.

Step 5: Validate + Publish Agent Card

5.1. Assemble full card:

{
  "name": "data-analysis-agent",
  "description": "Performs statistical analysis and visualization on tabular datasets.",
  "url": "https://agent.example.com",
  "version": "1.0.0",
  "provider": {
    "organization": "Example Corp",
    "url": "https://example.com"
  },
  "authentication": {
    "schemes": ["oauth2"],
    "credentials": { ... }
  },
  "capabilities": {
    "streaming": true,
    "pushNotifications": false,
    "stateTransitionHistory": true
  },
  "skills": [ ... ],
  "defaultInputModes": ["text/plain"],
  "defaultOutputModes": ["text/plain"]
}

5.2. Validate:

• Parse as JSON, check no syntax err
• All required fields present (name, description, url, skills)
• Each skill has id, name, description, min 1 I/O mode
• URL reachable, serves card at /.well-known/agent.json

5.3. Publish:

• Serve at https://<agent-url>/.well-known/agent.json
• Set Content-Type: application/json
• Enable CORS if cross-origin discovery needed
• Register with relevant agent registries

5.4. Test by fetching:

curl -s https://agent.example.com/.well-known/agent.json | python3 -m json.tool

Got: Valid JSON Agent Card at well-known URL. Any A2A client can parse.

If fail: JSON invalid? Use linter to find syntax err. URL unreachable? Check DNS, SSL cert, web server config. CORS needed? Add Access-Control-Allow-Origin headers.

Checks

• Agent Card valid JSON, no syntax err
• Required fields present: name, description, url, skills
• Each skill has id, name, description, inputModes, outputModes
• Auth scheme matches deploy security
• Capability flags match impl
• Served at /.well-known/agent.json, right Content-Type
• A2A clients fetch + parse OK
• Examples realistic, trigger right skill

Pitfalls

• Overpromising capabilities: streaming: true or pushNotifications: true without impl = client fails when used. Be conservative.
• Vague skill description: "does data stuff" blocks accurate matching. Be specific about inputs, outputs, domains.
• Missing CORS headers: Browser A2A clients can't fetch Agent Card without CORS.
• Skill overlap: Two skills handle same task → client can't pick. Keep boundaries clear.
• Forgetting default modes: No defaultInputModes/defaultOutputModes → clients unsure what content types to send.
• Version stagnation: Bump version when skills/capabilities change. Clients cache old versions.
• Publish before impl: Agent Card = contract. Publishing unimplemented skills → runtime failure.

See Also

• implement-a2a-server - impl server behind Agent Card
• test-a2a-interop - validate Agent Card conformance + interop
• build-custom-mcp-server - MCP as alt/complement to A2A
• configure-mcp-server - MCP config patterns for A2A setup