How theailanguage/a2a_samples works
Project type and purpose
• This is a sample-repo of four progressive implementations of Google’s A2A (Agent-to-Agent) protocol. Each version is a standalone microservice demo:
• version_1_simple: basic Flask HTTP client/server
• version_2_adk_agent: single ADK (Agent Development Kit) agent with JSON-RPC
• version_3_multi_agent: multi-agent orchestration (OrchestratorAgent + child agents + registry)
• version_4_multi_agent_mcp: distributed multi-agent + MCP (Model Context Protocol) integration
File structure
• Top-level “version_n…” directories each contain:
• agents/ (1 or more agent implementations)
• app/cmd/ (CLI entrypoint)
• client/ (client module or CLI)
• models/ (Pydantic or JSON-RPC schema)
• server/ (server and TaskManager)
• utilities/ (discovery, registry, MCP connectors in v3/v4)
• pyproject.toml (build/config)
README insights
• Each version README calls out its architecture:
• v1: simple REST endpoints
• v2: uses ADK runners & Gemini LLM under the hood
• v3: uses a registry (agent_registry.json) + OrchestratorAgent
• v4: adds MCP discovery/config (mcp_config.json) and dynamic MCP connectors
• Green for clients/CLI
• Blue for agent servers
• Orange for discovery/registry
• Purple for external LLM/MCP services
• Rectangles for microservices