Overview
An MCP server that bridges the Model Context Protocol (MCP) with the Google A2A (Agent-to-Agent) protocol. It enables MCP clients, such as Claude, to seamlessly discover, register, communicate with, and manage tasks on A2A agents via a unified interface. The server acts as an integration layer: a FastMCP-based API exposes tools for agent management, messaging, and task handling; an A2A client talks to registered A2A agents; a task manager forwards and tracks work; and an agent card fetcher retrieves agent metadata. It supports multiple transport types (stdio for local/desktop usage, streamable-http for web clients, and SSE for real-time updates) controlled by the MCP_TRANSPORT environment variable. Deployment paths include PyPI, Smithery, or local installation. Claude Desktop/Cursor IDE/Windsurf can connect to the MCP endpoint at http://host:port/mcp and use MCP tools like register_agent, list_agents, send_message, get_task_result, and cancel_task. This architecture emphasizes secure, composable interactions between MCP-compatible assistants and A2A agents through a single bridge server.
Features
Agent Management
Register, list, and unregister A2A agents with the bridge server.
Communication
Send messages to A2A agents and receive responses, with real-time streaming support.
Task Management
Track tasks, retrieve results by task_id, and cancel running tasks.
Transport Support
Supports stdio, streamable-http, and SSE transports with configurable MCP_TRANSPORT.
Who Is This For?
- MCP Client:Discover, register, and communicate with A2A agents through MCP tools.
- MCP Integrator:Integrates MCP with A2A to bridge protocols for tool developers.
- AI Assistants Vendors:Enable MCP-based assistants to access A2A agents via the bridge for interactive tasks.
