notebooklm-mcp-cli

# notebooklm-mcp-cli MCP server that bridges Claude Code to NotebookLM via browser session cookies. Exposes your NotebookLM notebooks as native Claude Code tools. **Transport**: stdio (local server, runs on your Windows machine) **Auth**: Playwright captures Google session cookies — no API key required --- ## Tools | Tool | Description | |------|-------------| | `notebooklm_auth` | Open Chrome, sign in to Google, store session cookies | | `notebooklm_list_notebooks` | List all notebooks in your account | | `notebooklm_get_notebook` | Get full detail: sources + notes for a notebook | | `notebooklm_query_notebook` | Ask a question against a specific notebook | | `notebooklm_extract_sources` | Extract all source content (for RAG ingestion) | | `notebooklm_search_all` | Search a query across all notebooks at once | --- ## Quick Start ```bash # Install npm install npx playwright install chromium # Build npm run build # Register in Claude Code (~/.claude/settings.json): # See docs/SETUP.md for the full JSON snippet # Then in Claude Code: # "Run notebooklm_auth to authenticate" # "List my NotebookLM notebooks" ``` --- ## Stack | Package | Purpose | |---------|---------| | `@modelcontextprotocol/sdk` | MCP server framework | | `playwright` | Browser automation for cookie capture | | `zod` | Input validation | | `typescript` | Strict mode TS | | `tsx` | Dev runner (no build step) | --- ## Cookie Storage Cookies are stored at `~/.notebooklm-mcp/auth.json` using XOR encryption with a machine-specific key (hostname). Expire after 6 days. **Never committed** — protected by `.gitignore` and the pre-commit hook. --- ## Register MCP in Claude Code Add to `~/.claude/settings.json`: ```json { "mcpServers": { "notebooklm": { "command": "node", "args": ["D:/AI_Dev/WorkspaceOS/Projects/notebooklm-mcp-cli/dist/index.js"], "env": {} } } } ``` --- ## Docs - [Full Setup Guide](docs/SETUP.md) --- ## Disclaimer NotebookLM has no public API. This server uses reverse-engineered browser endpoints. It may break if Google changes their internal API. For personal/research use only — not for production client deployments.