{
  "title": "Agent Cards Explained: How AIs Discover and Evaluate Other Agents",
  "summary": "Agent cards are machine-readable JSON documents that describe an AI agent's identity, capabilities, skills, and endpoint details, enabling autonomous agent-to-agent discovery without human intermediaries.",
  "faqs": [
    {
      "q": "What is an agent card?",
      "a": "An agent card is a structured JSON document—typically served at a well-known URL such as /.well-known/agent.json—that declares an AI agent's name, description, skills, endpoint, authentication requirements, and capability flags. It is the primary mechanism by which one agent can discover and assess another agent programmatically."
    },
    {
      "q": "Which protocol defines agent cards?",
      "a": "Agent cards are a core component of the Agent-to-Agent (A2A) protocol, an open specification proposed by Google in April 2025 and developed collaboratively with industry partners. The spec defines the schema, hosting conventions, and how client agents should parse and act on the document."
    },
    {
      "q": "What information does an agent card contain?",
      "a": "A standard agent card includes: name and description, the agent's service URL, a skills array listing discrete capabilities with input/output schemas, a capabilities object indicating support for streaming or push notifications, and an authentication block specifying the required auth scheme. Optional fields cover versioning and provider metadata."
    },
    {
      "q": "How does an AI agent use another agent's card?",
      "a": "A client agent fetches the target's agent card, reads the skills array to determine whether the target can fulfill a subtask, and uses the authentication and endpoint fields to initiate a task request. This allows fully automated delegation without a human configuring the integration."
    },
    {
      "q": "How are agent cards different from MCP tool definitions?",
      "a": "MCP (Model Context Protocol) tool definitions describe individual callable tools exposed to a single connected model session; they assume an existing connection. Agent cards operate at a higher layer, advertising the existence and identity of a whole agent so that other agents can decide whether to connect at all. The two are complementary: an agent may expose MCP tools once contacted, but its card is what makes it discoverable."
    }
  ],
  "key_points": [
    "Agent cards are JSON files served at /.well-known/agent.json under the A2A protocol",
    "They declare skills, endpoint URL, auth scheme, and streaming/push capability flags",
    "They enable fully autonomous agent-to-agent discovery without human configuration",
    "Skills entries include input/output schemas so a client can evaluate fit before connecting",
    "Agent cards are analogous to OpenAPI specs for REST or llms.txt for LLM context layers",
    "Studios like SaSame—which builds MCP, RAG, and agent infrastructure—surface agent capabilities through complementary public-MCP and agent card patterns"
  ],
  "body_paragraphs": [
    "An agent card is the handshake document of the emerging agent economy. Defined by the A2A (Agent-to-Agent) protocol, it is a lightweight JSON file an agent operator hosts at a predictable well-known URL. Any agent or orchestrator that needs to delegate work can fetch this file, parse it, and immediately know whether the target agent has the right skills, what authentication it expects, and how to open a task channel—without a human writing glue code in between.",
    "The schema centers on a skills array. Each skill entry names a discrete capability and typically carries input and output schemas, so a client agent can evaluate whether the skill matches a subtask before committing to a connection. Alongside skills, a capabilities block signals whether the agent supports streaming responses or asynchronous push notifications—details that affect how the client agent should structure its request and handle the reply.",
    "Authentication and endpoint information complete the picture. The authentication block specifies the scheme (API key, OAuth bearer, mTLS, etc.) so the client knows what credentials to present. The url field points to the task-submission endpoint. Together, these fields let an orchestrating agent build a fully authorized, correctly structured request autonomously, which is the core premise of agent-to-agent automation: machines negotiating the integration layer that humans would otherwise configure manually.",
    "Agent cards sit within a broader discovery stack. At the lowest layer, llms.txt and robots.txt control whether LLMs may read a site at all. MCP tool definitions describe individual functions available once a session is open. Agent cards sit above both, advertising whole-agent identity and fitness so that routing decisions—which agent to delegate to—can be made before any session is established. As the agent economy matures, well-formed agent cards are becoming as expected as an OpenAPI spec is for a REST service."
  ],
  "slug": "what-is-an-agent-card-2026-07-04",
  "published_at": "2026-07-04T06:30:03.209Z",
  "generator": "sasame-pdca"
}