Skip to main content

Build distributed systems without coupling behavior to the wire

MAGPIE is a transport-agnostic messaging and RPC framework for developers and AI agents. It gives applications the same small communication API whether data travels over ZeroMQ, MQTT, WebRTC, or a transport you add yourself.

The central idea is simple:

Separate where intelligence runs from where sensing and action happen.

An AI agent can run in the cloud, a perception pipeline on an edge computer, and a motor service on a robot. MAGPIE connects them without forcing application code to become networking code.

What MAGPIE provides​

DATA FLOWTopic-based streaming

Publish telemetry, events, images, audio, and typed frames without coupling producers to consumers.

COMMANDSRequest and response RPC

Call remote services with acknowledgements, timeouts, correlation, and transport-independent handlers.

CONTRACTSSchema-based APIs

Describe methods with JSON Schema and use the same contract for validation, dispatch, and discovery.

AI NATIVEMCP integration

Expose MAGPIE services directly as tools that AI clients can discover and call.

Why teams choose it​

  • One mental model. StreamWriter, StreamReader, RpcRequester, and RpcResponder remain recognizable across transports and languages.
  • Real interoperability. Python, C++, and TypeScript implementations share the same wire formats and can communicate directly.
  • Deployment freedom. Use direct local connections, brokered communication across NAT, or peer-to-peer WebRTC without redesigning the application.
  • Optional complexity. The core stays small. MQTT, WebRTC, media, discovery, and MCP are enabled only when needed.
  • Useful operational tools. Command-line readers, writers, request clients, media utilities, network discovery, and SSH-over-MQTT are part of the ecosystem.

Choose your path​

If this is your first time here, follow these three pages:

  1. Choose a language for the runtime and capabilities you need.
  2. Install MAGPIE with only the optional features your project uses.
  3. Run the quickstart and send your first message.

If you already know messaging systems, start with the four core APIs or compare the available transports.

:::tip The documentation is concept-first Most guides use synchronized Python, C++, and TypeScript tabs. Choose a language once and Docusaurus remembers it while you move through the site. :::