jupyter_mcp_sandboxes is an optional extension for jupyter-mcp-server that adds sandbox-backed code execution. It is discovered automatically through the jupyter_mcp_server.extensions entry point (powered by reactor) once installed.

Features

Installing this extension exposes four additional MCP tools:

• launch_sandbox — launch a code sandbox (eval, docker, jupyter-server, datalayer, daytona, e2b, coreweave, cloudflare, google-colab, kaggle, monty, modal).
• list_sandboxes — list launched sandboxes and their state.
• use_sandbox — select (or clear) the active sandbox used by execute_code.
• terminate_sandbox — stop and unregister a sandbox.

It also lets SANDBOX_VARIANT route the standard notebook/cell execution tools through a non-jupyter backend via the sandbox's plain kernel client when the selected variant exposes one.

Installation

pip install jupyter_mcp_sandboxes

Backend-specific extras are provided by the underlying code-sandboxes package, e.g.:

pip install "code-sandboxes[modal]"       # Modal
pip install "code-sandboxes"              # Google Colab (included in base)
pip install "code-sandboxes[kaggle]"      # Kaggle
pip install "code-sandboxes[daytona]"     # Daytona
pip install "code-sandboxes[e2b]"         # E2B
pip install "code-sandboxes[coreweave]"   # CoreWeave
pip install "code-sandboxes[cloudflare]"  # Cloudflare

Usage

Once installed, start jupyter-mcp-server as usual — no flag is required. The sandbox tools appear automatically. To route execute_code through a specific backend by default, set SANDBOX_VARIANT:

SANDBOX_VARIANT=monty jupyter-mcp-server start --transport streamable-http

License

BSD 3-Clause License. Copyright (c) 2024- Datalayer, Inc.