What is A2A Protocol?
A2A Protocol (Agent-to-Agent Protocol) is an open interoperability standard initiated by Google that enables AI agents built on different frameworks to discover, communicate, and collaborate with each other through a unified interface.
Quick Facts
| Full Name | Agent-to-Agent Protocol |
|---|---|
| Created | 2025 by Google |
| Specification | Official Specification |
How It Works
The A2A Protocol was introduced by Google in April 2025 as a complementary standard to MCP (Model Context Protocol). While MCP focuses on connecting models to tools and data sources, A2A addresses the challenge of inter-agent communication — allowing agents from different vendors and frameworks to work together on complex tasks. The protocol is built on HTTP and JSON-RPC, using familiar web standards for broad compatibility. By 2026, A2A has been adopted by major cloud providers and enterprise AI platforms, establishing itself alongside MCP as a foundational protocol for the agentic AI ecosystem.
Key Characteristics
- Agent Cards — JSON metadata files that describe an agent's capabilities, skills, and endpoint for discovery
- Task-based interaction model — agents collaborate by creating, updating, and completing shared tasks
- Multi-turn conversations — supports extended dialogues between agents with context preservation
- Push notifications via SSE/webhooks — enables asynchronous, event-driven agent collaboration
- Framework-agnostic — works across LangChain, CrewAI, Eino, AutoGen, and other agent frameworks
- Built on HTTP + JSON-RPC — leverages existing web infrastructure without proprietary protocols
Common Use Cases
- Enterprise workflow automation where specialized agents from different departments need to collaborate
- Multi-vendor AI ecosystems where agents from Google, Microsoft, and other providers must interoperate
- Complex task decomposition where a planning agent delegates sub-tasks to specialized execution agents
- Cross-organization agent collaboration in supply chain, healthcare, or financial services
- Agent marketplace platforms where third-party agents offer services to other agents
Example
Loading code...Frequently Asked Questions
What is the difference between A2A and MCP?
MCP (Model Context Protocol) connects AI models to tools and data sources, enabling a single agent to access external capabilities. A2A (Agent-to-Agent Protocol) connects multiple agents to each other, enabling collaboration between independently built agents. They are complementary: MCP handles vertical integration (agent-to-tool), while A2A handles horizontal integration (agent-to-agent).
How does agent discovery work in A2A?
A2A uses Agent Cards — JSON metadata files hosted at a well-known URL (typically /.well-known/agent.json). Each card describes the agent's name, capabilities, supported skills, authentication requirements, and endpoint URL. Client agents fetch these cards to discover what a remote agent can do before initiating collaboration.
Is A2A Protocol open source?
Yes, A2A is an open protocol released under the Apache 2.0 license. Google initiated the project and contributed it to open governance. The specification, reference implementations, and SDKs are all publicly available, and contributions from other organizations are welcomed.
Which agent frameworks support A2A?
As of 2026, A2A is supported by major frameworks including LangChain/LangGraph, CrewAI, AutoGen, Google ADK, Eino, and Semantic Kernel. Most modern agent orchestration platforms provide built-in A2A client and server capabilities.
What transport protocols does A2A use?
A2A is built on standard HTTP with JSON-RPC 2.0 message format. For real-time streaming, it supports Server-Sent Events (SSE). For asynchronous notifications, it uses webhooks. This design leverages existing web infrastructure and requires no proprietary transport layer.