NotebookLM Mindmap Extractor

Compile a NotebookLM mindmap and its generated responses into a hierarchical Markdown document using Playwright and Chrome DevTools Protocol (CDP).

Install

Requires Python 3.10 or newer, Playwright 1.64.1 or newer, and Chrome.

pip install notebooklm-mindmap-extractor

Or from a checkout:

python -m venv .venv
# Windows PowerShell
.\.venv\Scripts\Activate.ps1
# macOS/Linux: source .venv/bin/activate
python -m pip install -e ".[dev]"

For an existing wheel, use python -m pip install dist/<wheel-filename>.whl. The package installs the notebooklm-mindmap-extractor command. The module entry point is python -m notebooklm_mindmap_extractor.

Prepare Chrome

Launch Chrome with a dedicated data directory and a local debugging port. Sign in to Google manually in that browser, open your notebook, and open the target mindmap. The extractor attaches to this session; it does not automate sign-in.

Windows PowerShell (adjust the executable path if needed):

& 'C:\Program Files (x86)\Google\Chrome\Application\chrome.exe' --remote-debugging-port=9222 --remote-debugging-address=127.0.0.1 --user-data-dir="$env:LOCALAPPDATA\NotebookLMMindmapChrome"

macOS:

'/Applications/Google Chrome.app/Contents/MacOS/Google Chrome' --remote-debugging-port=9222 --user-data-dir="$HOME/.notebooklm-mindmap-chrome"

Linux:

google-chrome --remote-debugging-port=9222 --user-data-dir="$HOME/.notebooklm-mindmap-chrome"

Chrome 136 and newer require a non-default data directory for remote debugging. See Chrome's remote debugging guidance. Keep the endpoint local and exclude profiles and private exports from source control.

Extract

With the target mindmap already open:

notebooklm-mindmap-extractor --output output/mindmap.md

Select a particular notebook and open a map by its exact visible title:

notebooklm-mindmap-extractor --notebook "Token Buffet" --mindmap-title "Token MindMap" --max-nodes 1 --output output/token-smoke.md

Both notebook.google.com and notebooklm.google.com notebook links, IDs, and titles are accepted.

Two steps: capture, then clean

Extraction runs in two steps so you never have to re-extract to adjust the output.

1. Capture saves each node's text exactly as the page shows it, with no processing, to <run>/tmp/. The .txt files mirror the folder layout of the final .md files (Overview.txt for the root and for nodes with children, <title>.txt for leaves).
2. Clean (cleaner.py) reads those files and writes the .md files next to tmp/. It removes citations, page icons and the follow-up prompt, turns tables into Markdown tables, fences box diagrams, and separates paragraphs.

A normal run does both. To do them separately:

# capture only
notebooklm-mindmap-extractor --notebook "My Notebook" --mindmap-title "My Map" --extract-only

# clean from the saved raw files, as often as needed, with no browser
notebooklm-mindmap-extractor --clean handbook/my-notebook/my-map-20261028_152843

The page text carries no markup, so bold, bullets and heading markers are not available; headings and list items arrive as plain paragraphs. --build only combines the cleaned .md files into one document and does no cleaning.

Recompiling from Nodes and Customizing with doc_full_order.yml

Extraction automatically saves a doc_full_order.yml file inside your versioned run directory under --folder-output. It establishes the parent and child hierarchy with explicit order: attributes:

- nodeTitle: "Token Buffet AI Subscription Hub"
  childNodes:
    - title: "Core Architecture"
      order: 1
    - title: "Smart Routing & Tiers"
      order: 2

- nodeTitle: "Core Architecture"
  childNodes:
    - title: "Proxy Runtime: Standard HTTP server"
      order: 1
    - title: "Local Router: Port 8090"
      order: 2

You can edit this file to change order numbers, reorder items, or omit sections.

To recompile the unified document offline using a relative path, without launching Chrome or NotebookLM:

python -m notebooklm_mindmap_extractor --build handbook/token-buffet-ai-subscription-hub-20261027_175222

Or using a custom outline file:

python -m notebooklm_mindmap_extractor --build handbook/token-buffet-ai-subscription-hub-20261027_175222 --outline custom_order.yml --output output/custom_mindmap.md

Extraction clicks nodes and triggers NotebookLM generation. A missing or stale response causes a nonzero exit instead of a success placeholder. The output is written after all requested nodes complete.

Development and tests

python -m pip install -e ".[dev]"
python -m playwright install chromium
python -m pytest
python -m build
python -m twine check dist/*

Tests cover hierarchy reconstruction, the cleaner, and the capture-then-clean pipeline on a saved real extraction (tests/fixtures/run/). They do not contact Google or need a browser.

python -m build creates a source archive and wheel in dist/. Packaging follows the Python Packaging User Guide. The package is MIT licensed (see LICENSE). Releases are published to PyPI by .github/workflows/publish.yml when a GitHub Release is published, using Trusted Publishing.

Project layout

pyproject.toml                       Package metadata, dependencies, console entry point
src/notebooklm_mindmap_extractor/
    __init__.py                     Public data model and hierarchy helper
    __main__.py                     python -m entry point
    cli.py                          Arguments and extraction orchestration
    extractor.py                    Raw capture to tmp/, then the clean step per node
    cleaner.py                      Cleaning of the raw page text
    builder.py                      doc_full_order.yml and combining the .md files
    hierarchy.py                    Parent-child reconstruction
    models.py                       MindmapNode dataclass
tests/                              Unit tests and a saved real extraction (fixtures/run)

The former root script is replaced by the installed console and module commands. requirements.txt remains an editable-install shortcut.

Current limitations

The extractor supports mindmap trees rendered directly in the notebook or inside the embedded Studio App viewer. Wait for the viewer to finish loading before starting. Node clicks occur inside the map frame while generated answers are captured from the notebook chat panel. The command leaves the signed-in browser open and never reloads the notebook.

NotebookLM selectors may need updates when Google's UI changes. Only rendered nodes are discovered; collapsed branches are not automatically expanded. Complete accessibility levels use DOM preorder. Coordinate inference assumes a left-to-right map and uses the nearest parent vertically, so inspect the resulting hierarchy. Markdown headings are capped at six levels. A live test against a signed-in notebook is separate from the local fixture tests.