Jupyter MCP Server

pip install jupyter-mcp-server

Datalayer's server connects to a running Jupyter server and gives the model the notebook operations a person would use: switch between notebooks, read cells with or without outputs, insert, move, delete, and execute with multimodal results so a plot comes back as an image rather than a repr string. edit_cell_source is the detail that matters in practice — it applies find-and-replace edits inside a cell instead of overwriting the whole source, which stops a long analysis cell from being rewritten in full on every small correction. Execution does not have to stay on the local kernel either: launch_sandbox and use_sandbox route execute_code to a Docker, E2B, Daytona, Modal, Colab or Cloudflare backend, so the same tool call can run against a GPU box without changing the notebook.