Elastic introduces the Agent2Agent (A2A) communication standard, designed for multi-agent collaboration: agents do not need to share memory or tools, and can complete task splitting, state synchronization, and result feedback through JSON-RPC messages only. The official has open-sourced a "digital newsroom" example on Elasticsearch - six agents, including news director, reporter, researcher, archive, editor, and publisher, each perform their own duties and complete the entire process of topic selection, writing, review, and release through A2A collaboration.

image.png

Four Principles of the Protocol  

1. Message Passing: Structured Task & Artifact objects, supporting SSE streaming updates  

2. Coordination: Agent Card's automatic discovery capability, dynamic allocation of dependencies  

3. Specialization: Each agent focuses only on a single area, enabling horizontal scalability  

4. Distributed State: Task progress and intermediate results are broadcast in real-time, without a central scheduler

Relationship with MCP  

A2A is responsible for "multi-agent division of labor and collaboration", while MCP is responsible for "single-agent tool invocation"; the two can be used together: the reporter agent accesses the image library through MCP and submits the manuscript to the editor agent through A2A, achieving a "tool + collaboration" hybrid architecture.

Implementation Progress  

Currently, A2A has been integrated into Google ADK, LangGraph, and Crew.AI. Elastic provides an Agent Builder technical preview, allowing developers to start an "AI newsroom" sandbox with one click using Docker-Compose, and experience the fully automated process of topic selection → data → article writing → publishing within 15 minutes.