# A2A Protocol — Agent-to-Agent Communication Standard

> Google's open protocol for AI agents to discover, communicate, and collaborate across frameworks. SDKs for Python, TypeScript, Java, Go, and .NET. 23,000+ stars, Apache 2.0.

## Install

Save in your project root:

## Quick Use

**Python:**
```bash
pip install a2a-sdk
```

**TypeScript/JavaScript:**
```bash
npm install @a2a-js/sdk
```

**Go:**
```bash
go get github.com/a2aproject/a2a-go
```

**Java:** Available via Maven. **.NET:** `dotnet add package A2A`

Read the full specification: [a2a-protocol.org](https://a2a-protocol.org/latest/)

---

## Intro

A2A (Agent-to-Agent) is an open protocol by Google enabling AI agents built on different frameworks to discover capabilities, negotiate interaction formats, and collaborate securely — without exposing internal logic. With 23,000+ GitHub stars and Apache 2.0 license, it's now hosted by the Linux Foundation's Agentic AI Foundation. While MCP standardizes how agents connect to tools, A2A standardizes how agents talk to each other. SDKs are available for Python, TypeScript, Java, Go, and .NET, with 40+ official sample implementations.

Best for: developers building multi-agent systems where agents from different vendors need to communicate.
Works with: any agent framework (LangChain, CrewAI, AutoGen, Claude Agent SDK, etc.).
Complements: MCP (tool integration) — A2A handles agent-to-agent collaboration.

---

## A2A Protocol — Architecture & Concepts

### Core Concepts

#### Agent Cards
Every A2A agent publishes an "Agent Card" — a JSON document describing its capabilities, supported interaction modes, and authentication requirements. Other agents discover and evaluate capabilities through these cards.

```json
{
  "name": "Research Agent",
  "description": "Performs web research and synthesizes reports",
  "capabilities": ["web_search", "summarization", "citation"],
  "interaction_modes": ["sync", "streaming", "async"],
  "authentication": { "type": "oauth2" }
}
```

#### Communication Modes

| Mode | Use Case |
|------|----------|
| **Synchronous** | Quick request-response (< 30s) |
| **Streaming** | Real-time progressive results |
| **Asynchronous** | Long-running tasks with callbacks |

#### Rich Data Exchange
Agents can exchange text, files, structured JSON, and binary data — not just text messages.

### How A2A Differs from MCP

| Aspect | MCP | A2A |
|--------|-----|-----|
| **Purpose** | Agent ↔ Tool integration | Agent ↔ Agent communication |
| **Analogy** | USB port for peripherals | Language for conversations |
| **Discovery** | Tool manifests | Agent Cards |
| **Interaction** | Request/response | Sync, streaming, async |
| **State** | Stateless tool calls | Stateful task management |

**They complement each other:** An agent uses MCP to call tools and A2A to collaborate with other agents.

### Official SDKs

| Language | Package | Install |
|----------|---------|---------|
| Python | a2a-sdk | `pip install a2a-sdk` |
| TypeScript | @a2a-js/sdk | `npm install @a2a-js/sdk` |
| Go | a2a-go | `go get github.com/a2aproject/a2a-go` |
| Java | a2a-java | Maven |
| .NET | A2A | `dotnet add package A2A` |

### Sample Implementations

40+ official samples available across all languages, including:
- Multi-agent research systems
- Customer support agent collaboration
- Code review agent pipelines
- Financial analysis agent networks

### FAQ

**Q: What is the A2A Protocol?**
A: An open standard by Google (now Linux Foundation) that enables AI agents from different frameworks to discover each other, communicate via JSON-RPC 2.0, and collaborate on tasks — without exposing internal logic.

**Q: Is A2A free?**
A: Yes, Apache 2.0 license. All SDKs and specifications are free and open-source.

**Q: How does A2A relate to MCP?**
A: They're complementary. MCP handles agent-to-tool integration (like a USB port), while A2A handles agent-to-agent communication (like a shared language). Use both together for full agent interoperability.

---

## Source & Thanks

> Created by [Google](https://github.com/a2aproject), hosted by Linux Foundation. Licensed under Apache 2.0.
>
> [A2A](https://github.com/a2aproject/A2A) — ⭐ 23,000+

Thank you to Google and the A2A project team for building the definitive agent interoperability standard.

---

<!-- ZH -->

## 快速使用

**Python:** `pip install a2a-sdk`
**TypeScript:** `npm install @a2a-js/sdk`
**Go:** `go get github.com/a2aproject/a2a-go`

完整规范：[a2a-protocol.org](https://a2a-protocol.org/latest/)

---

## 简介

A2A（Agent-to-Agent）是 Google 发布的开放协议，让不同框架构建的 AI Agent 能够发现能力、协商交互格式并安全协作。拥有 23,000+ GitHub Star 和 Apache 2.0 许可证，现由 Linux 基金会托管。MCP 标准化 Agent 与工具的连接，A2A 标准化 Agent 之间的通信。提供 Python、TypeScript、Java、Go 和 .NET 的 SDK，40+ 官方示例。

适合人群：构建多 Agent 系统的开发者，需要不同厂商的 Agent 相互通信。
与 MCP 的关系：互补——MCP 管工具集成，A2A 管 Agent 间协作。

---

## A2A vs MCP

| 方面 | MCP | A2A |
|------|-----|-----|
| 用途 | Agent ↔ 工具 | Agent ↔ Agent |
| 类比 | USB 接口 | 共同语言 |
| 发现机制 | 工具清单 | Agent Card |

### FAQ

**Q: A2A 是什么？**
A: Google 发布的 Agent 间通信开放标准，让不同框架的 Agent 能发现、通信和协作。

**Q: 免费吗？**
A: Apache 2.0 完全免费开源。

---

## 来源与感谢

> Created by [Google](https://github.com/a2aproject). Licensed under Apache 2.0.
>
> [A2A](https://github.com/a2aproject/A2A) — ⭐ 23,000+

---
Source: https://tokrepo.com/en/workflows/0c013210-cbd4-4cb8-a5f7-8a77635698b2
Author: Agent Toolkit