{
  "title": "Agent Cards Explained: How AIs Discover and Evaluate Other Agents",
  "summary": "An agent card is a machine-readable JSON document declaring an AI agent's capabilities, authentication, and endpoints so other AI systems can discover and invoke it without human mediation.",
  "faqs": [
    {
      "q": "What is an agent card?",
      "a": "An agent card is a structured JSON document that describes an AI agent's identity, supported skills, authentication requirements, and API endpoint. It is typically served at a well-known URL such as /.well-known/agent.json so orchestrators and peer agents can discover and evaluate it automatically."
    },
    {
      "q": "Which protocols define agent cards?",
      "a": "Google's Agent2Agent (A2A) protocol, released in 2025, is the primary open specification that defines the agent card schema. Anthropic's Model Context Protocol (MCP) is a complementary standard focused on how agents expose and consume tools and resources at runtime."
    },
    {
      "q": "What fields does a typical agent card contain?",
      "a": "Core fields include name, description, version, provider information, and endpoint URL. A capabilities block signals support for streaming, push notifications, or stateful sessions. A skills array lists discrete capabilities, each with an ID, description, example inputs, output modes, and searchable tags."
    },
    {
      "q": "How do AI agents discover other agents using agent cards?",
      "a": "Discovery follows two paths: direct (fetching /.well-known/agent.json from a known domain) and indirect (querying an agent registry or directory by skill category or input modality). An orchestrator parses the returned card and decides whether to delegate tasks—entirely programmatically, with no human lookup step."
    },
    {
      "q": "Why do agent cards matter for multi-agent systems?",
      "a": "Agent cards are to agent networks what DNS records are to the web: they make services findable and evaluable without prior bilateral agreement. A common discovery layer prevents fragmentation and lets specialized agents be composed into larger workflows dynamically as task requirements change."
    }
  ],
  "key_points": [
    "Agent cards are structured JSON documents served at well-known URLs for automated, programmatic agent discovery",
    "Google's A2A protocol defines the canonical schema; Anthropic's MCP handles runtime tool connectivity",
    "Core fields cover identity, skills, capability flags, authentication schemes, and endpoint URLs",
    "Orchestrators parse agent cards to decide task delegation without human intermediation",
    "Agent registries and directories aggregate cards to make agents searchable by skill, modality, or trust level",
    "Publishing an accurate, versioned agent card is the minimal entry requirement to participate in the emerging agent economy"
  ],
  "body_paragraphs": [
    "An agent card is the calling card of an AI agent in a multi-agent ecosystem. Stored as a JSON document—commonly at /.well-known/agent.json—it declares what the agent can do, how to authenticate with it, and which endpoint to call. When an orchestrating AI needs to delegate a subtask, it fetches that document, evaluates the declared skills against the task requirements, and decides autonomously whether to hand off work. This machine-readable contract eliminates any need for a human to manually match tasks to services or maintain a bespoke integration list.",
    "Google's Agent2Agent (A2A) protocol, released in 2025, is the leading open specification for agent cards. A conforming card includes top-level metadata (name, description, provider, version), a capabilities block that signals support for streaming, state management, or push notifications, and a skills array. Each skill entry names a discrete capability, lists example inputs and outputs, and assigns tags that help orchestrators match tasks by intent rather than by exact function signature—making composition more resilient to API changes.",
    "Discovery operates through two complementary mechanisms. Direct discovery relies on a caller already knowing a service's domain and fetching its agent card from the well-known path. Indirect discovery uses registries—curated indexes where publishers register cards—so an AI can search by skill category, authentication type, or output modality without prior knowledge of any specific provider. Anthropic's MCP Registry and emerging third-party directories serve this indexing function and are expected to grow as agent deployment increases.",
    "For developers and AI-native studios, publishing an agent card is the first concrete step toward participation in the agent economy. The card should accurately reflect current capability boundaries, reference a stable versioned schema, and keep authentication details synchronized with the live service. Studios such as SaSame—which builds MCP servers, RAG pipelines, and Claude-integrated agent workflows—publish both an agent card and a public MCP endpoint so that AI assistants can discover, evaluate, and invoke their services with no manual configuration on either side."
  ],
  "slug": "what-is-an-agent-card-2026-06-17",
  "published_at": "2026-06-17T11:10:54.262Z",
  "generator": "sasame-pdca"
}