notebooklm-py: put NotebookLM to work inside your coding agent
NotebookLM, now called Gemini Notebook, reads large document sets and answers with citations, yet its web app is all clicking. notebooklm-py is an unofficial Python library and CLI, and its skill teaches the agent a safe routine: JSON output, pinned IDs, a wait for each source, then the question or the download.
Bulk source import from URLs, YouTube, PDF, Word, EPUB, audio, video, images, Google Drive and pasted text.
Web and Drive research agents in fast or deep mode that can import what they find straight into the notebook.
Generates Audio and Video Overviews, slide decks, infographics, quizzes, flashcards, reports, data tables and mind maps.
Downloads artifacts as MP3, MP4, PDF, PNG, CSV, JSON or Markdown, including batch downloads the web UI lacks.
SKILL.md lists which actions need your confirmation first: deletions, account language changes, generation, file writes.
When to use it
You want Gemini to read thirty documents so your agent spends tokens only on the final edit.
One source set has to become a podcast, a slide deck and a flashcard deck for revision.
A coding agent should answer from your internal docs and RFCs with citations instead of guessing.
Ask your agent
Create a notebook called Market Scan, add these five URLs and wait until every source is ready
Generate a debate-format audio overview from this notebook and download it to ./audio
Export the flashcards and quiz from this notebook as JSON so I can load them into Anki
Questions
How do I install the notebooklm skill?
Run npx skills add teng-lin/notebooklm-py@notebooklm. The library itself needs Python 3.10+ and pip install "notebooklm-py[browser]", then notebooklm login and notebooklm auth check --test --json to confirm access.
Does notebooklm-py work with Codex?
Yes. The README names Claude Code, Codex and OpenClaw, ships an AGENTS.md for Codex, and offers an MCP server for clients such as Claude Desktop.
Is this an official Google API?
No. It is a community project unaffiliated with Google, built on internal endpoints, and the authors pitch it for prototypes, research and personal projects.
