A2A Protocol Technical Documentation
Protocol Introduction

Protocol Introduction

The A2A (Agent2Agent) protocol is an open protocol initiated by Google Cloud, aimed at promoting interoperability between different AI Agents. Its core goal is to enable these agents—regardless of whether they are built by different vendors or use different technology frameworks—to communicate and collaborate effectively within a dynamic, multi-agent ecosystem.

As enterprises increasingly build and deploy autonomous agents to scale, automate, and enhance workflows (from ordering new laptops to assisting customer service representatives, to supply chain planning), collaboration between agents becomes crucial. The A2A protocol enables agents, even those built by different vendors or using different frameworks, to interoperate, thereby increasing autonomy, multiplying productivity gains, and reducing long-term costs.

The release of the A2A protocol is supported and contributed to by over 50 technology partners (such as Atlassian, Box, Cohere, Langchain, MongoDB, PayPal, Salesforce, SAP, ServiceNow, etc.) and leading service providers (such as Accenture, Deloitte, Infosys, KPMG, PwC, etc.). This collaborative effort reflects a shared vision: a future where AI agents, regardless of their underlying technology, can seamlessly collaborate to automate complex enterprise workflows, driving unprecedented efficiency and innovation.

A2A is an open protocol that complements Anthropic’s Model Context Protocol (MCP), which provides useful tools and context for agents. Drawing on Google’s internal expertise in scaling agent systems, the A2A protocol aims to address the challenges encountered when deploying large-scale multi-agent systems for customers.

A2A’s Design Principles

The design of the A2A protocol follows these key principles to enhance collaboration between agents, ensure flexibility, security, and compatibility with existing systems:

  1. Embrace agentic capabilities: Allows agents to collaborate in their natural, unstructured patterns without needing shared memory, tools, or context, enabling true multi-agent scenarios.
  2. Build on existing standards: The protocol is built on widely accepted technical standards (like HTTP, SSE, JSON-RPC), facilitating integration with existing enterprise IT stacks.
  3. Secure by default: Designed to support enterprise-grade authentication and authorization, ensuring only authorized users and systems can access agents.
  4. Support for long-running tasks: Flexibly supports various scenarios, from quick tasks to complex research that might take hours or even days (with human intervention), and can provide real-time feedback, notifications, and status updates during task execution.
  5. Modality agnostic: Supports multiple interaction forms, including text, audio, video streams, forms, iframes, etc., enhancing agent interaction capabilities and adaptability.

Participants in A2A

The A2A protocol defines three core participants:

  • User: The human or service using the agent system to accomplish a task.
  • Client: The entity representing the user that requests actions from the agent (service, application).
  • Server: The opaque (black-box) remote agent providing the service, i.e., the A2A server.

The following diagram clearly shows the positions of these three participants:

Note: Compared to the MCP protocol, the A2A protocol does not explicitly define a “Host” participant. This is a difference in design philosophy; A2A focuses more on open implementation and handles issues like security and discovery through other means (like AgentCard and standard protocols).

Core Concepts of A2A

The A2A protocol is built around several core concepts to enable effective interaction between agents:

  • AgentCard: A JSON file describing an Agent’s capabilities, typically hosted at the /.well-known/agent.json path, making it easy for clients to discover and understand the Agent.
  • Task: A stateful entity representing the process of collaboration between a client and a remote agent to achieve a specific outcome. Tasks contain status, history, and generated artifacts.
  • Artifact: An immutable output generated by the agent as the final result of a task, which can contain multiple Parts.
  • Message: Used to pass non-artifact content between the client and the agent, such as instructions, context, thought processes, status updates, etc.
  • Part: An atomic unit of content within a message or artifact, having a specific content type (e.g., text, file, form).

These concepts together form the foundation of the A2A protocol, enabling different agents to communicate and collaborate in a structured, standardized way. Subsequent sections will introduce these core concepts in more detail.

Technical Architecture

Core Components

  1. Agent Card

    • Public metadata file in JSON format
    • Describes agent capabilities and authentication requirements
    • Supports automated capability discovery
    • Provides standardized service descriptions

  2. A2A Server

    • Implements standard protocol endpoints
    • Handles task requests and execution
    • Manages task lifecycles
    • Supports real-time status updates

  3. A2A Client

    • Provides standardized interfaces
    • Sends and manages tasks
    • Handles asynchronous responses
    • Supports streaming data transmission

Communication Model

  1. Task Management

    • Task-centric communication model
    • Tasks have a complete lifecycle
    • Supports immediate completion or long-running operations

  2. Capability Discovery

    • Broadcasts capabilities via AgentCard
    • Supports dynamic service discovery
    • Flexible negotiation mechanisms

  3. User Experience Negotiation

    • Standardized message formats
    • Support for multiple content types
    • Flexible UI capability negotiation

Application Scenarios

  1. Intelligent Customer Service

    • Multi-turn conversation management
    • Knowledge base integration
    • Automated ticket processing

  2. Development Collaboration

    • Code review assistance
    • Documentation automation
    • Test case generation

  3. Data Analysis

    • Data cleaning and transformation
    • Visual report generation
    • Anomaly detection and alerting

Related Resources

💡 Technical Support:

Quick Start