Jupyter Mcp Server

What is Jupyter MCP Server?

Jupyter MCP Server is a service based on the Model Context Protocol (MCP) that gives AI tools in advanced IDEs and CLI tools the ability to connect and manage Jupyter Notebooks for tasks like data analysis, visualization, and machine learning. It supports multiple Notebooks simultaneously, interactive execution, and multimodal output (text, images, tables).

How to use Jupyter MCP Server?

Install via uvx with uvx better-jupyter-mcp-server or clone the source and run uv run fastmcp run src/better_jupyter_mcp_server/server.py. You must have a running Jupyter Server (e.g., jupyter lab --port 8888 --IdentityProvider.token YOUR_TOKEN) and provide the server URL and token in your client’s rules file. Then configure the MCP JSON with "command": "uvx" and "args": ["better-jupyter-mcp-server"] using stdio transport.

Key features of Jupyter MCP Server

• MCP compatible: works in any IDE or CLI tool supporting MCP
• Multi‑Notebook management: manage several Notebooks at once
• Interactive execution: adapts strategy based on cell output
• Multimodal output: returns text, images, tables, and more
• Tools for notebook, cell, and advanced integrated operations

Use cases of Jupyter MCP Server

• Data analysis and visualization with AI assistance across multiple Notebooks
• Machine learning workflow management (cleaning, feature engineering, training, evaluation)
• Interactive debugging of code snippets via temporary cells without saving to Notebook
• Automating repetitive data science tasks by combining notebook and cell tools
• Leveraging multimodal LLMs (like Gemini 2.5) for rich output interpretation

FAQ from Jupyter MCP Server

What makes Jupyter MCP Server different from other Notebook tools?

Unlike tools that only read/edit or handle a single Notebook, it manages multiple Notebooks interactively, supports multimodal output, and provides advanced integrated cells like append_execute_cell to reduce tool calls.

What are the dependencies and runtime requirements?

Python 3.12+, uv for installation, and a running Jupyter Server (Lab or Notebook) with a token. The server uses stdio transport and runs locally.

How does authentication work?

You provide the Jupyter server URL and token in your client’s rules file. The server then uses that token to authenticate all calls.

Are there any known limits or performance notes?

The connect_notebook tool can take 10–30 seconds because it starts a Kernel. Other tools execute quickly. Multimodal output requires a client that supports returning image data via MCP.

Where does data live?

Notebooks and kernels run on the configured Jupyter Server (default localhost:8888). All data stays on that server; the MCP server does not store data externally.