Skip to main content
Agent discovery enables clients to find and understand agent capabilities through standardized metadata. The AgentCard serves as the agent’s “business card” containing identity, skills, and operational requirements.

AgentIdentity

Schema:
Use Case: Decentralized Agent Identity
What it’s for: Establishing verifiable decentralized identity for agents using DIDs. Enables trustless discovery and authentication without central authorities. The DID document contains public keys for cryptographic verification.

AgentInterface

Schema:
Use Case: Multiple Transport Support
What it’s for: Declaring multiple communication protocols an agent supports. Allows clients to choose between JSON-RPC, WebSocket, gRPC, or other transports based on their needs.

AgentExtension

Schema:
Use Case: AP2 Payment Extension
What it’s for: Declaring support for protocol extensions like AP2 (payments), custom authentication schemes, or specialized capabilities. The required flag indicates if clients must support the extension.

Skill

Schema:
Use Case: Data Analysis Skill
What it’s for: Defining what an agent can do. Supports both simple inline definitions and rich Claude-style documentation with SKILL.md files. Orchestrators use skills to match agents to tasks.

AgentCapabilities

Schema:
Use Case: Full-Featured Agent
What it’s for: Declaring which optional A2A protocol features the agent supports. Clients can choose agents based on required capabilities like streaming or push notifications.

AgentCard

Schema:
Use Case: Complete Agent Card
What it’s for: The complete agent metadata package. Clients fetch the AgentCard to discover capabilities, skills, security requirements, and operational characteristics before interacting with an agent.

Agent Discovery Flow

Summary: Agent discovery uses AgentCard as the central metadata structure containing identity (AgentIdentity), capabilities (AgentCapabilities), skills (Skill), and operational requirements. Agents declare support for extensions (AgentExtension) and multiple transports (AgentInterface). Clients discover agents through registries, evaluate their capabilities and costs, then establish secure connections using the declared security schemes and trust configurations.