notebook-mcp 0.1.0

MCP server that provides notebook-aware context and analysis for .ipynb files

notebook-mcp

MCP server that provides notebook-aware context and analysis for .ipynb files.

Features

• Offline notebook analysis (no kernel required)

  • Strip outputs (optional)
  • Stable cell indexing
  • AST-based defines / uses
  • Best-effort dependency graph
  • Focused context slicing for a target cell
  • Export notebook to a deterministic Python script (best-effort)

• Optional Jupyter Server adapter (HTTP)

  • List active sessions
  • Inspect kernel metadata

Install (dev)

python -m pip install -e ".[dev]"

Distribution

Publish Python package (PyPI)

1. Build:

python -m pip install -e ".[dev]"
python -m build

2. Upload:

python -m twine upload dist/*

Publish NPM wrapper

The Node wrapper lives in npm/ and publishes as a scoped package.

cd npm
npm publish --access public

Run (stdio transport; best for IDE agents)

notebook-mcp

Run (streamable HTTP transport)

set MCP_TRANSPORT=streamable-http
set MCP_HOST=127.0.0.1
set MCP_PORT=8000
notebook-mcp

Jupyter Server adapter

Set env vars:

• JUPYTER_BASE_URL e.g. http://127.0.0.1:8888
• JUPYTER_TOKEN (if required)

Notes

• Dependency graph and stale detection are best-effort offline. Truthful execution/runtime state requires kernel messaging.