FastA2A
A framework-agnostic Python implementation of the Agent2Agent protocol that turns any agent into an A2A server, with separate storage, broker, and worker components so the task-executing worker can run on its own machine or process.
FastA2A exposes an agent as an A2A-compatible ASGI application built on Starlette, with request and response validation through Pydantic and tracing through OpenTelemetry. It is designed so the worker that actually executes tasks can live outside the web server, on a different machine or in a separate process, while in-memory storage and broker implementations are provided to get started quickly; a real deployment implements its own worker component for its agentic framework of choice.
A streaming endpoint answers with server-sent events: a task record first, then status changes and artifacts as the worker publishes them, ending once the task reaches a final state.
The library originated inside Pydantic AI and was moved to its own repository after community interest, and still ships a Pydantic AI integration that turns any agent into an A2A app with one function call; other agentic frameworks are invited to add their own worker implementations.