NotebookLM Skill: Query NotebookLM from Claude Code

What It Is

NotebookLM Skill is a 7.6k-star Claude Code skill that lets your coding agent chat directly with Google NotebookLM — the source-grounded knowledge base powered by Gemini. The official README describes it in one line: “Use this skill to query your Google NotebookLM notebooks directly from Claude Code for source-grounded, citation-backed answers from Gemini. Browser automation, library management, persistent auth. Drastically reduced hallucinations — answers only from your uploaded documents.”

It solves the problem every developer hits when they tell an agent to “search through my local documentation”: massive token consumption from re-reading files, keyword-matching that misses context, hallucinations when the agent can’t find something and invents plausible-sounding APIs, and the endless copy-paste dance between NotebookLM in the browser and your editor. With this skill, Claude asks NotebookLM directly, Gemini synthesizes an answer from your uploaded sources, and Claude writes the code — no manual switching, no invented APIs.

The mechanics are simple. When you mention NotebookLM or paste a notebook URL, Claude loads the skill’s instructions, runs its Python scripts, opens a real Chrome browser (via Patchright, a Playwright fork), asks your question, and returns the answer straight in the CLI. One-time Google authentication persists across sessions, and your notebook library — names, topics, tags — is stored locally in the skill folder as library.json.

What Makes It Different

The core difference is source grounding. Every answer comes exclusively from the documents you uploaded, which is why the README calls hallucinations “drastically reduced”: if the information isn’t in your notebook, the model says so instead of inventing content. That’s a fundamentally different behavior from feeding docs into context or letting an agent web-search.

Second, it’s NotebookLM rather than local RAG, and the README makes the trade-off explicit. Local RAG costs hours of setup — embeddings, chunking, a vector database — and still has retrieval gaps. NotebookLM does the preprocessing server-side: upload docs once, get Gemini-synthesized answers with citations, multi-source correlation across 50+ documents, and zero infrastructure on your machine.

Third, it’s a skill, not an MCP server — and the author maintains both. The skill version is Python, stateless (fresh browser per question), and Claude-Code-only, designed for quick integration; the MCP server version is TypeScript, keeps persistent chat sessions, and works with Codex, Cursor, and other tools. Each answer even ends with “Is that ALL you need to know?” to prompt Claude into comprehensive follow-ups — a small touch that turns single queries into real research.

Who It’s For

NotebookLM Skill fits three types of users. First, Claude Code users who already keep docs in NotebookLM — API docs, workshop manuals, spec sheets. The README’s flagship example is querying a Suzuki GSR 600 workshop manual for brake fluid type, oil specs, and torque values, with Claude returning DOT 4, SAE 10W-40, and 100 N·m with citations.

Second, developers building against their own documentation. The second official example is building an n8n workflow for Gmail spam filtering using an “n8n notebook” — Claude asks the notebook comprehensive questions, then produces a working workflow on the first try without debugging hallucinated APIs.

Third, anyone who wants RAG-quality answers without RAG infrastructure. If you’ve been avoiding local RAG because of the embedding/chunking/vector-DB setup tax, this gives you the payoff in about five minutes: clone, authenticate once, upload docs, query.

Skip it if you use Claude in the web UI (skills run in a sandbox without network access, so this won’t work there), or if you need persistent multi-turn sessions — that’s what the MCP server is for. Also note the free-tier NotebookLM rate limits and the requirement that notebooks be shared with anyone with the link. It’s MIT licensed.

Bottom Line

NotebookLM Skill turns Google’s source-grounded knowledge base into a back-end for Claude Code: upload your documents once, then ask questions directly from the terminal and get citation-backed answers with minimal hallucination risk. The install is a single git clone into ~/.claude/skills, authentication takes one Chrome session, and the payoff is no more copy-paste between browser and editor. If you work with manuals, API docs, or research collections, it pairs naturally with the Kreuzberg/Xberg document intelligence skill for extraction on one side and this for grounded Q&A on the other — or with anything-to-notebooklm to get documents into a notebook in the first place. For broader research workflows, see our AI search & research FAQ.