What does this skill do, and when should you use it?
This is a Claude Code Skill that uses Patchright browser automation to let local Claude Code talk to Google NotebookLM directly. For each question it opens a fresh browser session, retrieves a source-grounded, citation-backed answer from Gemini based solely on your uploaded documents, then closes. It also provides notebook library management (saving, searching and activating notebook links with metadata) and one-time persistent Google authentication. It works only with local Claude Code, not the web UI.
The skill runs Python scripts through a python scripts/run.py wrapper: auth_manager.py opens a visible Chrome window for manual Google login and persists session state; notebook_manager.py saves notebook URLs with name, description and topics into a local library., with list, search, activate, remove and stats commands; ask_question.py opens a Patchright-driven browser session, types your question into the specified NotebookLM notebook, retrieves the Gemini answer with source citations, and appends a follow-up prompt ("Is that ALL you need to know?") that pushes Claude to ask supplementary questions. All data (auth state, browser cookies, library) lives in ~/.claude/skills/notebooklm/data/ protected by .gitignore. The virtual environment and Chrome install automatically on first run.
- A developer who has uploaded their documentation library (API docs, framework manuals) to NotebookLM and wants Claude Code to query it instead of repeatedly reading files — saving tokens and reducing hallucinations.
- A mechanic or technician querying a workshop manual, e.g. asking the Suzuki GSR 600 manual for brake fluid type, oil specs and axle torque, and getting exact in-document answers.
- An n8n user who wants Claude to build a working workflow on the first try from their own notebook instead of guessing APIs.
- A user with multiple notebooks who wants Claude to automatically pick the right one for a task via the topic-tagged library.
- An existing user of PleasePrompto/notebooklm-mcp who prefers a clone-and-go Python skill without running an MCP server.
How do you install this skill?
- Where the persisted login session is stored and how to clear it isn't documented
- Shell / CLI
- Network access
- Local filesystem
Python 3.8+Patchright 1.55.2Google ChromeGoogle account
In a terminal:
mkdir -p ~/.claude/skillscd ~/.claude/skillsgit clone https://github.com/PleasePrompto/notebooklm-skill notebooklm- Open Claude Code and ask: "What are my skills?"
On first use, the skill automatically creates a .venv, installs dependencies and Chrome. Note: local Claude Code only — the web UI sandbox has no network access and cannot run it. No other installation methods are documented.
git clone --depth 1 https://github.com/PleasePrompto/notebooklm-skill.git ~/.claude/skills/notebooklm-skillGenerated from the source repository and skill path; it copies only this skill's folder. If the author's install steps above differ, follow those first. To scope it to one project, replace ~/.claude/skills with that project's .claude/skills.
How do you use this skill?
Once installed, send your agent any of these to trigger it:
- Set up NotebookLM authentication
- Add [link] to my NotebookLM library
- What does my React docs say about hooks?
- Say "Set up NotebookLM authentication" in Claude Code; a Chrome window opens for manual Google login (browser must be visible; one-time only).
- Create a notebook at notebooklm.google.com, upload documents (PDFs, Google Docs, websites, YouTube, etc.), and copy the link via ⚙️ Share → Anyone with link.
- Say "Add [link] to my NotebookLM library" (Claude can use Smart Add — querying the notebook to discover its content and metadata — or you supply name/description/topics manually).
- Ask questions directly, e.g. "What does my React docs say about hooks?" — Claude selects the notebook and returns grounded answers; if information is incomplete, it asks follow-ups as prompted, then synthesizes.
What are this skill's strengths and limitations?
- Answers come strictly from your uploaded documents with citations, drastically reducing hallucinations.
- Direct CLI Q&A eliminates the copy-paste dance between browser and editor.
- Smart library management: save, search and auto-select notebooks by topic tags.
- One-time authentication persists across sessions; everything is self-contained in the skill folder with no global installs.
- Minimal setup — clone and go, with dependencies installed automatically.
- Local Claude Code only — the web UI sandbox has no network access, so it cannot run there at all.
- No session persistence: each question opens a new browser and cannot reference prior answers, and browser overhead adds seconds per question.
- Rate limits of roughly 50 queries per day on free Google accounts.
- Relies on browser automation with human-like patterns to evade detection; the README explicitly cannot guarantee Google won't detect or flag it and recommends a dedicated account.
- Documents must be manually uploaded to NotebookLM and notebooks must be shared publicly; no test suite is evident in the repository.
How does this skill compare with similar options?
Side by side with related skills; every score comes from the same FSRS standard.
| Skill | FS score | Stars | Last updated | License |
|---|---|---|---|---|
| NotebookLM Research Assistant Skill this page | 63 · Recommended | ★ 7.8k | 1mo ago | MIT |
| Web App Testing Skill | 49 · Use with care | ★ 3.1k | 8mo ago | MIT |
| NotebookLM Automation | 53 · Use with care | ★ 20k | 11d ago | MIT |
| Azure Content Understanding for Python ✓ Microsoft · Official | 48 · Use with care | ★ 2.7k | 3mo ago | MIT |
| Playwright Interactive Debugger ✓ OpenAI · Official | 39 · Not recommended | ★ 28k | 4mo ago | — |
How did FollowSkills review this skill?
Depends on browser automation against the NotebookLM web UI; page structure changes could break it, and the doc offers no stability evidence or tests.
Requires manual Google login through a visible browser window (not silent credential harvesting) — reasonably transparent design, but the doc doesn't specify where the persisted session/auth is stored or how to clear it.
Open a dimension to read why it scored that way
Evidence confidence：Low — Mostly static review, author material or a limited demo; useful for discovery, not high-risk decisions.
This review was completed under FSRS 1.0. Dimension scores are not converted; it is queued for re-review under FSRS 2.0.
See the full review method →