NotebookLM gets an unofficial API, cookies included

What it does

notebooklm-mcp-cli wraps Google NotebookLM’s internal endpoints in a single Python package that exposes both a terminal CLI (nlm) and an MCP server. You can script notebook creation, ingest sources from URLs or Drive, generate audio and video artifacts, and run batch or cross-notebook queries. It also plugs into Claude, Cursor, Gemini, and other AI assistants so they can manipulate your research data directly.

The interesting bit

The project sidesteps the lack of an official API by automating browser cookie extraction from Chrome, Arc, Brave, and others, then uses those credentials to talk to NotebookLM’s undocumented internals. It also auto-generates MCP configuration for a half-dozen editors, sparing you the usual JSON-editing drudgery.

Key highlights

• Dual personality: one install provides both the nlm CLI and the notebooklm-mcp server.
• 35 MCP tools covering notebooks, sources, studio content, slide revisions, research, and pipelines.
• Auto-detects and injects MCP server config into Claude Code, Cursor, Gemini CLI, Windsurf, and more.
• Supports named browser profiles for multi-account Google authentication.
• Ships with a doctor diagnostic command and optional “AI skills” files to teach agents how to use the tools.

Caveats

• Relies entirely on undocumented internal APIs that Google can change or block without warning.
• Authentication requires extracting cookies from your browser; the README notes this is mandatory and provides a tool for it.
• The MCP surface is large—35 tools can consume significant context window space, and the README explicitly warns users to disable the server when it is not needed.

Verdict

Grab it if you want to automate NotebookLM workflows or let an AI agent manage your research. Avoid it if you need stability guarantees, official support, or a clean auth flow that doesn’t involve rummaging through your browser’s cookie jar.