Jupyter MCP Server is a mature, well-documented open-source MCP server maintained by Datalayer, focused on letting AI agents operate Jupyter Notebooks in real time. Its strengths are multimodal output, multi-notebook support, execution that adapts from cell output feedback, and the ability to route execution to a dozen-plus cloud sandbox backends. The main caveats are that agents can execute arbitrary code, so tokens and network exposure must be managed carefully, and that code-sandboxes plus the mcp SDK have strict version pairing requirements that should be checked before upgrading.
Jupyter MCP Server is an MCP server developed by Datalayer (BSD 3-Clause licensed) that lets AI clients connect to and manage Jupyter Notebooks in real time. It offers live view of notebook changes, smart execution that adjusts from cell output feedback, full notebook context awareness, multimodal output support (images, plots, text), and multi-notebook switching. It works with any MCP client (Claude Desktop, Cursor, Windsurf, and more) and with any Jupyter deployment such as a local JupyterLab or JupyterHub. Through the optional jupyter_mcp_sandboxes extension, code execution can be routed to sandbox engines including Datalayer, Kaggle, Google Colab, Modal, Daytona, E2B, CoreWeave and Cloudflare. The project also ships JupyterLab integration (enabled by default), a jupyter-cite prompt, OAuth 2.1 sign-in, and OpenTelemetry-based observability.
{
"mcpServers": {
"jupyter": {
"command": "uvx",
"args": ["jupyter-mcp-server@latest"],
"env": {
"JUPYTER_URL": "http://localhost:8888",
"JUPYTER_TOKEN": "MY_TOKEN",
"ALLOW_IMG_OUTPUT": "true"
}
}
}
}