Skill Detail

Edit and verify a scoped Jupyter notebook task through MCP

Connect an external MCP agent to a selected Jupyter notebook, make bounded cell edits, execute the affected analysis and return saved results with execution evidence.

Developer ToolsMCP
Developer Tools MCP Security Reviewed
โญ 1.3k GitHub stars
COPY SKILL INSTRUCTIONS (OPTIONAL)
npx skills add agentskillexchange/skills --skill edit-verify-scoped-jupyter-notebook-task-mcp Copy
Uses the third-party skills CLI, not an ASE-owned installer. Check your agent’s compatibility. This copies instructions; complete the upstream tool setup below separately.
At a glance
Tools required
Python >=3.10; uv/uvx; JupyterLab; jupyter-collaboration; jupyter-mcp-tools; ipykernel; MCP client
Install & setup
In the notebook environment install jupyterlab jupyter-collaboration jupyter-mcp-tools ipykernel. Start JupyterLab with token authentication on loopback port 8888 and confirm collaboration autosave works. Configure the MCP client command uvx with args ["jupyter-mcp-server==2.2.5"], env JUPYTER_URL=http://127.0.0.1:8888, JUPYTER_TOKEN set privately to the server token, and DOCUMENT_ID set to the authorized relative notebook path. Use a compatible Python environment; v2 requires MCP SDK >=2,=1.1.1. Keep credentials out of version control.
Author
Datalayer
Publisher
Community
Last updated
Nov 13, 2026
Quick brief

Use Jupyter MCP Server when an external assistant needs to perform a bounded analysis or repair in an existing Jupyter notebook while preserving notebook and kernel context. Input the authorized notebook path, allowed cell changes, data locations, expected outputs and execution budget. Return the updated notebook, changed-cell summary, observed outputs and unresolved execution errors.

How it works

What this skill actually does

Prepare a JupyterLab environment with collaboration support and a suitable kernel. Configure the MCP server with the authorized JUPYTER_URL, JUPYTER_TOKEN and a relative DOCUMENT_ID for the selected notebook. Connect over stdio and inspect the tools exposed by the installed release. Read the notebook and relevant cells before editing. Use the notebook and cell tools to make the agreed change, execute the affected cells in dependency order, and inspect returned text, plots and errors. Re-read changed cells and confirm the notebook has saved before reporting completion.

If execution fails, retain the error and relevant kernel context, make only a justified repair within scope, and report what was rerun. State when success depends on previous kernel state. Do not claim a clean-kernel rerun unless it was actually performed. Ask the operator before restarting a shared kernel or running additional costly work.

Invoke this when the active agent lives outside JupyterLab and needs direct, state-aware cell operations rather than manually copying code and outputs between chat and a notebook. The boundary is one selected notebook task with a reviewable execution handoff. Installing an in-notebook chat assistant, launching a cloud sandbox fleet, publishing results and managing arbitrary server files are separate workflows.

MCP is the invocation framework. The v2.2.5 README supports the local stdio setup, notebook/cell operations and execution feedback. Package provenance was checked against PyPI; runtime execution and security were not tested during intake.