jupyter-mcp-server

Jupyter MCP Server, developed by Datalayer, lets an AI agent connect to and manage Jupyter notebooks in real time. The README lists key features: real-time control with instant view of changes, smart execution that adjusts when a cell fails using the cell output, awareness of the whole notebook, multimodal output (images, plots, text), multiple notebooks, JupyterLab integration, compatibility with MCP clients such as Claude Desktop, Cursor and Windsurf, and a hook system with OpenTelemetry tracing.

It works with any Jupyter you already run, local or JupyterHub, with no account needed, and installs with `pip install jupyter-mcp-server` (there is also a Docker image). Code can run on different sandbox variants chosen with `--sandbox-variant`: jupyter, datalayer, kaggle, colab, monty or modal. Notebook documents and code execution are separate choices; as of v1.3.2 the `--provider` flag is renamed `--document-provider`, with the old name still accepted as a deprecated alias.

The README also promotes Datalayer's hosted endpoint, a Claude Code plugin and OAuth sign-in with scopes (`notebooks:read`, `notebooks:write`, `code:execute`, `data:read`); those are the vendor's service, while the open-source server is separate. The license is BSD 3-Clause and it was last pushed on 2026-10-11. This review is based on the README and repository data; we have not run it ourselves.