Elastic has released the Agent2Agent (A2A) open protocol

Elastic has introduced Agent2Agent (A2A), a communication standard designed specifically for multi-agent collaboration. With A2A, agents can decompose tasks, synchronize states, and return results solely through JSON-RPC messages, without needing to share memory or tools. Elastic has already open-sourced a “Digital Newsroom” reference implementation on Elasticsearch—featuring six specialized agents (News Editor, Reporter, Researcher, Archivist, Copy Editor, and Publisher) that coordinate via A2A to autonomously manage the entire workflow from story ideation and drafting to review and publication.

Four Core Principles of the A2A Protocol

1. Message Passing: Structured Task and Artifact objects with support for Server-Sent Events (SSE) for streaming updates.
2. Coordination: Agents automatically discover each other’s capabilities via Agent Cards and dynamically assign task dependencies.
3. Specialization: Each agent focuses on a single domain, enabling horizontal scaling.
4. Distributed State: Task progress and intermediate results are broadcast in real time—no central orchestrator required.

Relationship with MCP

• A2A handles multi-agent coordination and delegation.
• MCP (Model Calling Protocol) handles tool invocation within a single agent.

The two protocols are complementary and can be used together: for example, a Reporter agent might use MCP to access an image library while using A2A to submit a draft to the Editor agent—enabling a hybrid “tools + collaboration” architecture.

Adoption & Implementation Status

A2A is already integrated into Google ADK, LangGraph, and Crew.AI. Elastic also provides a technical preview of Agent Builder, allowing developers to launch an “AI Newsroom” sandbox with a single docker-compose command and experience a fully automated pipeline—from topic selection and data gathering to article generation and publishing—in under 15 minutes.