Claude Code borrows NotebookLM’s citation-backed memory

What it does

This is a Claude Code Skill that automates Chrome to query your Google NotebookLM notebooks directly from the terminal. It returns answers synthesized by Gemini that are grounded exclusively in your uploaded documents, complete with citations, so Claude can write code or explain concepts without reading entire file trees. A local JSON library tracks your notebooks so Claude can pick the right one for the task.

The interesting bit

Instead of building a local RAG pipeline with vector databases and chunking strategies, it piggybacks on NotebookLM’s pre-processed knowledge base—trading infrastructure for browser automation. Each query spins up a fresh browser session, asks the question, and immediately nudges Claude with “Is that ALL you need to know?” to trigger deeper follow-ups.

Key highlights

• Source-grounded answers only from uploaded docs, with explicit citations.
• Self-contained Python skill with auto-managed Chrome and isolated dependencies.
• Persistent local auth and notebook library management across sessions.
• Human-like typing and interaction patterns to avoid bot detection.
• Stateless per-query browser sessions; no persistent chat context inside NotebookLM.

Caveats

• Works only with local Claude Code; the web UI sandbox blocks network access.
• Each question opens a fresh browser, so you can’t reference “the previous answer.”
• Notebooks must be shared publicly via link, and the free tier carries daily query limits.

Verdict Worth a look if you already live in Claude Code and want citation-backed answers without maintaining a local vector database. Skip it if you need strict privacy (docs live on Google’s servers) or if you do most of your work in the Claude web UI.