nteract

nteract is a local-first notebook environment where humans, kernels, and AI agents can work with the same live document. It ships as a native desktop app with realtime sync across windows and programmatic clients, and managed local environments.

Built on jupyter-zmq-client (opens in a new tab) and jupyter-protocol (opens in a new tab).

What's in here

Runtime model

nteract stores notebook content in an Automerge-backed document and keeps live kernel state in an explicit runtime state document. The daemon owns kernel processes, execution queues, output capture, and the write path for execution results.

Execution requests name a synced cell_id; the daemon reads the cell source from the shared document, records queue and status changes in runtime state, and writes outputs through the same model. The desktop app, CLI, MCP server, and agent clients use that path, so programmatic control observes the same notebook state a human is editing.

MCP Server

Run nteract mcp to connect AI assistants to notebooks. Local notebooks use the runtime daemon; direct hosted connections use WebSockets without a local daemon. Agents can run code, read and write cells, manage dependencies, and collaborate with humans on the same live notebook.

Usage

List open notebooks with kernel and environment info:

Project structure

Prerequisites

Linux only: Install GTK/WebKit dev libraries:

Common commands

Library crates

The underlying Rust libraries are published to crates.io:

• `jupyter-protocol` (opens in a new tab): Jupyter messaging protocol
• `jupyter-zmq-client` (opens in a new tab): Jupyter kernel interactions over ZeroMQ
• `nbformat` (opens in a new tab): Notebook parsing