a2aproject/a2a

Agent2Agent (A2A) is an open communication protocol that lets AI agents built by different companies, using different frameworks, and running on separate servers talk to each other and collaborate. Think of it like a common language for AI agents: without a shared standard, an AI agent built with one toolkit cannot work with an agent built with another, even if they could theoretically complement each other. The core challenge A2A addresses is that modern AI systems often need multiple specialized agents working together, one agent might be good at researching information, another at scheduling, another at writing. A2A gives these agents a standard way to discover each other's capabilities, negotiate what format they'll exchange data in (text, files, forms, etc.), and coordinate on long-running tasks, all without any agent needing to expose its internal workings or proprietary logic. Technically, A2A uses JSON-RPC 2.0 over HTTP as its transport layer, supports both real-time streaming and asynchronous communication, and includes a concept called "Agent Cards", structured descriptions that agents publish so other agents can discover what they can do. It is designed for enterprise use with security and authentication built in. You would use A2A when building multi-agent AI systems where different agents, potentially from different vendors, need to collaborate. SDKs are available in Python, Go, JavaScript, Java, and .NET.