Skip to content
Per-cell access control. Grant or lock the agent's write access, cell by cell. 
Live presence. The status bar shows what the agent is doing as it happens. 
Inline review. Comment on any cell; the agent replies in the same thread.
Your notebook is already in the browser. Now your agent can be too.
A JupyterLab and JupyterLite extension that lets a browser agent read, edit, run and review the live notebook through WebMCP. No server, no API keys.
The agent proposes a one-line fix inline. The diff is reviewable before it sticks: same cell, same kernel, same tab.
install
bash
pip install jupyterlite-webmcp
jupyter labPublished on PyPI. The extension only does something in a browser that exposes document.modelContext; elsewhere it stays out of the way. Full install guide.
what it does
- the live notebook
- Unsaved edits, your exact text selection, the running kernel and its outputs. State that exists only inside the tab. Why it matters.
- 22 webmcp tools
- Read, navigate, edit, execute and review through
document.modelContext. Writes are guarded by a source hash, so the human always wins. Tool reference. - you set the limits
- Per-cell and per-notebook access levels, plus Propose mode where each edit waits for your Accept or Deny. Propose mode.
- visible by design
- Presence rings, state badges, inline diffs and output provenance show every agent action in the notebook itself.
- review threads
- Comments live in the notebook metadata, so the conversation travels with the .ipynb.
- one pip install
- A prebuilt extension for JupyterLab, Notebook 7 and JupyterLite. No Node.js, no lab build. Install.
screenshots
built for the OpenAI WebMCP Challenge
Selected as one of the 10 winners. Created by Allison Coleman (@alliecatowo) with Juan Mendoza (@mennymendoza). See where it's going on the roadmap, and tell us what you want an agent to do in your notebook in Discussions.