xtralab-skills

xtralab

An opinionated JupyterLab meta-package for use with coding agents.

It bundles a curated set of JupyterLab extensions, a path-first file browser, git diffs rendered by xtralab, an agent launcher, and a quieter default workspace.

Install

pip install xtralab

Usage

As a JupyterLab package

Once installed, start JupyterLab the usual way:

jupyter lab

As a desktop app

A standalone Electron build is also available, packaged as a DMG on macOS and an AppImage on Linux. Each tagged release ships installers on the releases page, and builds from main are uploaded as workflow artifacts on the Actions tab. See CONTRIBUTING.md for the desktop app architecture and local build instructions.

What's included

xtralab builds on ajlab, the agent-ready JupyterLab base. On top of it, xtralab pulls in JupyterLab 4.6+ with jupyterlab-git, jupyterlab-lsp, jupyterlab-quickopen, jupyterlab-vim, and a set of light and dark themes (jupyterlab-cursor-light, jupyterlab-cursor-dark, jupyterlab-day, jupyterlab-night). See pyproject.toml for the full list and pinned versions.

The bundled xtralab labextension then adds:

• A path-first file browser in the left sidebar.
• Git diffs rendered by xtralab in the jupyterlab-git panel, using @pierre/diffs for text and notebooks and a 2-up / swipe / onion-skin view for images.
• An agent launcher with a prompt box, buttons for the agents installed on your machine (Claude, Codex, Antigravity, Copilot, Goose, OpenCode, Kiro, Mistral Vibe), a collapsible list of changed files, and an Open row for a terminal, notebook, console, or your terminal editor (Neovim or Vim).
• A Terminals panel listing the running terminal sessions, each badged with the agent or editor detected inside it. Open terminal tabs in the main area carry the same icon.
• Sidebar toggle buttons in the top bar for the left and right areas.

Connecting agents to Jupyter (MCP)

xtralab runs a Model Context Protocol server inside JupyterLab, provided by jupyter-server-mcp. To let a coding agent drive JupyterLab through it, register the bundled jupyter-server-mcp-proxy console script with the agent:

claude mcp add jupyter -- jupyter-server-mcp-proxy
codex mcp add jupyter -- jupyter-server-mcp-proxy
copilot mcp add jupyter -- jupyter-server-mcp-proxy

Run this from a terminal inside xtralab, so the proxy inherits the server environment and discovers the running server automatically. No port is needed, and a single registration keeps working across restarts. The launcher also surfaces these commands with copy buttons, filtered to the agents you have installed.

See the jupyter-server-mcp README for other MCP clients.

Agent skill

xtralab ships an Agent Skills entry at agent-skill/ that teaches any coding agent how to customize JupyterLab from plain-English requests like "hide the status bar", "change the theme", or "add a language server". The same SKILL.md file works with Claude Code, Codex CLI, Gemini CLI, GitHub Copilot, and other tools that read the Agent Skills format.

For Claude Code, this repository doubles as a one-plugin marketplace. Inside a Claude Code session:

/plugin marketplace add jtpio/xtralab
/plugin install xtralab-skills@xtralab

For agents that read from ~/.agents/skills/, such as Codex CLI or Gemini CLI, clone and copy the skill directory:

git clone --depth=1 https://github.com/jtpio/xtralab.git /tmp/xtralab
mkdir -p ~/.agents/skills
cp -r /tmp/xtralab/agent-skill/skills/customize-jupyterlab ~/.agents/skills/

See agent-skill/README.md for additional install paths, the list of supported agents, and what the skill knows.

Language servers

xtralab ships [jupyterlab-lsp][lsp] with two language servers pre-registered:

• Python, using ty, which is bundled and works out of the box.

• TypeScript and JavaScript, using typescript-language-server, which you install yourself:

  npm install -g typescript-language-server typescript
  

  Restart JupyterLab afterwards to pick it up.