Quickstart
Get up and running with MCP Jupyter in minutes.
Requirements
- UV - Required for installation
- JupyterLab with
jupyter-collaboration
andipykernel
- An MCP-compatible client (e.g., Goose, Cursor)
Installation
MCP Jupyter Server uses stdio and can be added to any MCP client with:
uvx mcp-jupyter
Quick Setup
1. Start Jupyter Server
First, set up and start your Jupyter server:
# Create virtual environment
uv venv
source .venv/bin/activate
# Install dependencies
uv pip install jupyterlab jupyter-collaboration ipykernel
# Start Jupyter server
jupyter lab --port 8888 --IdentityProvider.token BLOCK --ip 0.0.0.0
tip
The server expects a token for authentication. If the TOKEN
environment variable is not set, it defaults to "BLOCK".
2. Configure Your MCP Client
For Goose
Add MCP Jupyter to your Goose configuration:
goose session --with-extension "uvx mcp-jupyter"
For Other Clients
Add the following to your MCP client configuration:
{
"mcpServers": {
"jupyter": {
"command": "uvx",
"args": ["mcp-jupyter"]
}
}
}
3. Start Using
Once configured, you can:
- Create or open a notebook through your AI assistant
- Execute code cells with preserved state
- Let the AI handle errors and install packages
- Switch between manual and AI-assisted work seamlessly
Example Session
# Your AI assistant can help you:
# 1. Load and explore data
# 2. Visualize results
# 3. Debug errors
# 4. Install missing packages
# All while preserving your notebook state!