Agent2Agent Protocol (A2A)
An open protocol enabling communication and interoperability between opaque agentic applications.
Overview
One of the biggest challenges in enterprise AI adoption is getting agents built on different frameworks and vendors to work together. That's why Google created an open Agent2Agent (A2A) protocol, a collaborative way to help agents across different ecosystems communicate with each other. Google is driving this open protocol initiative for the industry because they believe this protocol will be critical to support multi-agent communication by giving your agents a common language – irrespective of the framework or vendor they are built on. With A2A, agents can show each other their capabilities and negotiate how they will interact with users (via text, forms, or bidirectional audio/video) – all while working securely together.
Conceptual Overview
The Agent2Agent (A2A) protocol facilitates communication between independent AI agents. Here are the core concepts:
- Agent Card: A public metadata file (usually at
/.well-known/agent.json) describing an agent's capabilities, skills, endpoint URL, and authentication requirements. Clients use this for discovery.
- A2A Server: An agent exposing an HTTP endpoint that implements the A2A protocol methods (defined in the json specification). It receives requests and manages task execution.
- A2A Client: An application or another agent that consumes A2A services. It sends requests (like
tasks/send) to an A2A Server's URL.
- Task: The central unit of work. A client initiates a task by sending a message (
tasks/send or tasks/sendSubscribe). Tasks have unique IDs and progress through states (submitted, working, input-required, completed, failed, canceled).
- Message: Represents communication turns between the client (role: "user") and the agent (role: "agent"). Messages contain Parts.
- Part: The fundamental content unit within a Message or Artifact. Can be TextPart, FilePart (with inline bytes or a URI), or DataPart (for structured JSON, e.g., forms).
- Artifact: Represents outputs generated by the agent during a task (e.g., generated files, final structured data). Artifacts also contain Parts.
- Streaming: For long-running tasks, servers supporting the streaming capability can use
tasks/sendSubscribe. The client receives Server-Sent Events (SSE) containing TaskStatusUpdateEvent or TaskArtifactUpdateEvent messages, providing real-time progress.
- Push Notifications: Servers supporting pushNotifications can proactively send task updates to a client-provided webhook URL, configured via
tasks/pushNotification/set.
Typical Flow
- Discovery: Client fetches the Agent Card from the server's well-known URL.
- Initiation: Client sends a
tasks/send or tasks/sendSubscribe request containing the initial user message and a unique Task ID.
- Processing:
- (Streaming): Server sends SSE events (status updates, artifacts) as the task progresses.
- (Non-Streaming): Server processes the task synchronously and returns the final Task object in the response.
- Interaction (Optional): If the task enters input-required, the client sends subsequent messages using the same Task ID via
tasks/send or tasks/sendSubscribe.
- Completion: The task eventually reaches a terminal state (completed, failed, canceled).
Getting Started
- 📚 Read the technical documentation to understand the capabilities
- 📝 Review the json specification of the protocol structures
- 🎬 Use samples to see A2A in action:
- Sample A2A Client/Server (Python, JS)
- Multi-Agent Web App
- CLI (Python, JS)
- 🤖 Use sample agents to see how to bring A2A to agent frameworks:
- Agent Developer Kit (ADK)
- CrewAI
- LangGraph
- Genkit
- 📑 Review key topics to understand protocol details:
- A2A and MCP
- Agent Discovery
- Enterprise Ready
- Push Notifications
Contributing
Contributions are welcome! See the contributing guide to get started.
- Have questions? Join the community in GitHub discussions.
- Help with protocol improvement feedback in GitHub issues.
- Want to send private feedback? Use the Google feedback form.
What's Next
Future plans include improvements to the protocol itself and enhancements to the samples:
Protocol Enhancements:
- Agent Discovery:
- Formalize inclusion of authorization schemes and optional credentials directly within the AgentCard.
- Agent Collaboration:
- Investigate a QuerySkill() method for dynamically checking unsupported or unanticipated skills.
- Task Lifecycle & UX:
- Support for dynamic UX negotiation within a task (e.g., agent adding audio/video mid-conversation).
- Client Methods & Transport:
- Explore extending support to client-initiated methods (beyond task management).
- Improvements to streaming reliability and push notification mechanisms.
Sample & Documentation Enhancements:
- Simplify "Hello World" examples.
- Include additional examples of agents integrated with different frameworks or showcasing specific A2A features.
- Provide more comprehensive documentation for the common client/server libraries.
- Generate human-readable HTML documentation from the JSON Schema.
About
A2A Protocol is an open source project run by Google LLC, open to contributions from the entire community.
