The capability is real and broad — cross-notebook queries and pipelines are things the product's own UI does not offer — but the plumbing is undocumented Google endpoints and browser cookies, which can change without notice. Treat it as a power tool for personal and experimental work, which is the framing the README itself uses.
One Python package that is both a CLI (`nlm`) and an MCP server for Gemini Notebook. It uses Google's internal, undocumented APIs and browser cookies rather than a published API, which the README states plainly.
- Notebooks listed and created — `notebook_list`, `notebook_create`
- Sources added from a URL, plain text, Google Drive or a file, and Drive sources re-synced — `source_add`, `source_sync_drive`
- A notebook queried with the exchange persisting to the web UI, plus chat sessions listed, read and exported — `notebook_query`, `chat_list`, `chat_get`, `chat_export`
- Studio content generated — audio, video and more — and slide decks revised — `studio_create`, `studio_revise`
- Artifacts downloaded individually or in bulk across one notebook or all of them — `download_artifact`, `download_all_artifacts`
- Web and Drive research started from the agent — `research_start`
- Questions asked across several notebooks at once, multi-step pipelines, batch operations, and tag-based selection — `cross_notebook_query`, `pipeline`, `batch`, `tag`
- Notebook sharing, and a `nlm doctor` command for diagnosing setup problems
A Gemini Notebook account and cookies extracted from your browser — the package includes a tool for that. Installed from PyPI as `notebooklm-mcp-cli` 0.9.14, via uv, uvx, pip or pipx; the MCP entry point is `notebooklm-mcp`. `nlm setup add <tool>` writes the config for Claude Code, Claude Desktop, Gemini, GitHub Copilot, Cursor, Cline and Antigravity, or emits JSON for anything else. Claude Desktop users can double-click the `.mcpb` extension instead. Tested on Pro/free and Google AI Ultra tiers; Enterprise accounts are untested.
One command — uv tool install notebooklm-mcp-cli
