
An opinionated JupyterLab meta-package for coding agents.
xtralab reshapes JupyterLab around terminal coding agents. It bundles a curated
set of extensions, a path-first file browser, rich git diffs, an agent launcher,
a Model Context Protocol server that lets agents drive the app, and a quieter
set of defaults. It builds on ajlab, the
agent-ready JupyterLab base.

The documentation covers everything in detail: installation, a tour of the features, connecting agents over MCP, agent skills, and the desktop app.
Install
pip install xtralab
jupyter lab
A standalone desktop app (DMG on macOS, AppImage on Linux) ships with every tagged release on the releases page. The macOS app is signed and notarized and keeps itself up to date; see the desktop app docs for how it manages projects and kernels.
Highlights

The launcher starts any agent installed on your machine, with an optional prompt and a Changes list that jumps straight into a side-by-side git diff you can edit in place. Select code in an editor, a notebook, or a diff to send a prompt about it to a new or already-running agent, or queue prompts while you read. One omnibox searches files and commands, the Terminals panel shows which agent runs where, and quieter defaults keep the workspace calm. The documentation walks through each of these.
Connecting agents to JupyterLab (MCP)
xtralab runs a Model Context Protocol server inside JupyterLab,
provided by jupyter-server-mcp, so an agent can open files, run
cells, and read notebooks. Register the bundled proxy from a terminal inside
xtralab:
claude mcp add jupyter -- jupyter-server-mcp-proxy
The launcher surfaces the matching command for every agent you have installed, and the MCP docs page covers the details.
Contributing
See CONTRIBUTING.md for the development setup, the
desktop app architecture, and the documentation sources in docs/.
License
BSD-3-Clause