datalayer/jupyter-mcp-server

What it solves

Jupyter MCP Server allows AI agents to interact with Jupyter Notebooks in real-time. It bridges the gap between an AI's ability to write code and the ability to execute it within a persistent, interactive environment, allowing agents to manage notebooks, execute cells, and view multimodal outputs (like plots and images) directly.

How it works

It implements the Model Context Protocol (MCP), acting as a server that exposes a set of tools to an MCP-compatible client (such as Claude Desktop, Cursor, or Windsurf). The server connects to a Jupyter server (local, JupyterHub, or hosted via Datalayer) and provides tools for file management, notebook manipulation, and code execution. It can route execution to either a standard Jupyter kernel or a specialized "code sandbox" (e.g., Kaggle, Google Colab, Modal) for scaled execution.

Who it’s for

Data scientists, AI researchers, and developers who want to give their AI agents the ability to perform data analysis, create visualizations, and manage their Jupyter notebooks autonomously.

Highlights

• Real-time Control: Agents can view and modify notebook changes as they happen.
• Multimodal Support: Supports various output types, including images and plots.
• Flexible Execution: Supports multiple sandbox variants for code execution, including GPU-enabled environments.
• Comprehensive Toolset: Provides tools for reading, inserting, deleting, and executing cells, as well as managing kernels and sandboxes.
• Enterprise-Ready Security: Includes OAuth 2.1 sign-in for scoped access control without sharing passwords.
• MCP Compatibility: Works with any MCP-compatible client.