Introduction

The A2A protocol is an open standard for agent collaboration across diverse frameworks, guided by core principles:

• it embraces agents' natural, unstructured capabilities for true multi-agent scenarios without restricting them to mere tools;
• builds on widely-used standards like HTTP, SSE, and JSON-RPC for easy integration with existing IT stacks;
• ensures enterprise-grade security with authentication and authorization akin to OpenAPI;
• supports tasks ranging from quick actions to deep, human-involved long-running orchestrations spanning hours or days, delivering real-time feedback and updates;
• and remains modality-agnostic, accommodating text, audio, video streaming, and more to reflect the diverse agentic landscape.

Capabilities

A2A enables communication between a client agent, which formulates and sends tasks, and a remote agent, which acts to deliver accurate information or actions, through key capabilities:

• capability discovery, where agents use a JSON-based “Agent Card” to advertise skills, allowing the client to select and engage the best-suited remote agent;
• task management, where tasks, defined by the protocol, follow a lifecycle that supports immediate or long-running completion with ongoing status syncing and produce outputs called “artifacts”;
• collaboration, where agents exchange messages to share context, replies, artifacts, or user instructions;
• and user experience negotiation, where message “parts” (content like images) with specified formats enable agents to align on user interface needs, such as iframes, video, or web forms.

A2A ER Diagram

The AgentCard provides visibility into what the agent can do. It’s basically the descriptor used in a catalog for others to discovery. The Task is the runtime coordination element between agents.

Agent - Autonomous software entity with a unique identity that processes requests, produces responses, and communicates with other agents through standardized interfaces.