notebooklm-mcp

What it does

Automates Google NotebookLM through a real Chrome browser (via Patchright stealth mode with persistent fingerprinting), giving AI agents programmatic control over notebook chat, source ingestion, audio overview generation, and DOM-level citation extraction.

Who it's for

Research engineers and knowledge workers building AI pipelines that require live NotebookLM access—particularly those synthesizing documents at scale, generating audio summaries, or maintaining multiple research workspaces.

Common use cases

• Ingest research documents into a NotebookLM notebook via an AI agent, then query it and extract citations for multi-document synthesis
• Automatically generate audio overviews of ingested sources as part of a larger research or knowledge-management workflow
• Build multi-step analysis pipelines where Claude Code feeds sources to NotebookLM, processes responses, and tracks citations
• Maintain separate NotebookLM workspaces for different teams or projects using multi-account support within a single agent
• Run NotebookLM automation on headless servers (CI/CD, hosted agents) using HTTP transport instead of stdio

Setup pitfalls

• Initial authentication (setup_auth) requires a visible Chrome window for manual Google login; on headless Linux, run xvfb-run -a npx notebooklm-mcp once to establish the persistent profile, then subsequent runs operate fully headless
• Chrome (stable channel) must be installed—if it fails to launch, set BROWSER_CHANNEL=chromium to use the bundled Patchright Chromium instead
• WSL1 cannot launch Chromium and is not supported; WSL2+WSLg (Windows 11+) is fully supported
• The server persists browser profiles and library data to disk—ensure the host process has write permissions to the application data directory