How do I authenticate with NotebookLM?

If you see 401 Unauthorized or Invalid CSRF, run nlm login in your terminal to re-authenticate. The assistant will not fix auth directly.

Which actions require explicit confirmation?

Destructive actions like notebook_delete and source_delete, and generative actions like studio_create and source_sync_drive require confirm=True with explicit user consent.

How do I ensure grounded, complete answers?

Use notebook_query with clear goals and follow up to fill gaps. NotebookLM returns grounded results based on sources, and you should verify whether more information is needed.

notebooklm

Scanned

npx machina-cli add skill pablodiegoo/Data-Pro-Skill/notebooklm --openclaw

Files (1)

SKILL.md

2.9 KB

NotebookLM Assistant Skill

You have native Model Context Protocol (MCP) tools available to interact directly with the user's Google NotebookLM account. There is no need to run local python scripts or launch browsers manually—the notebooklm-mcp server handles everything natively.

Workflow

Verify Connection: You should already have access to tools starting with notebook_, source_, research_, studio_, etc. If any of these throw an authentication error (401 Unauthorized, Invalid CSRF), prompt the user to open their terminal and run nlm login to re-authenticate. DO NOT try to fix auth yourself.
Understand the Goal: Identify what the user wants to do: query an existing notebook, add sources, or generate artifacts (like audio overviews).
Execute via Native Tools: Call the appropriate MCP tools directly.

Detailed Tool Usage

To avoid cluttering your context, detailed lists of tools and their specific parameters have been moved to reference files. Read these only when you need them.

Available Tools & Capabilities: Read references/mcp_tools.md for a complete mapping of all 29 tools available to you (CRUD for notebooks, sources, notes, sharing, and studio artifacts). This includes important warnings about operations that require confirm=True.
Query Best Practices: Read references/best_practices.md if the user wants you to ask a question to a notebook via notebook_query. It contains critical instructions on how to handle NotebookLM's grounded responses and when to ask follow-up questions to synthesize a complete answer.

Key Principles

Confirmation Required: Destructive actions (notebook_delete, source_delete) and generative actions (studio_create, source_sync_drive) require you to pass confirm=True. You must get explicit user consent before doing this.
Asynchronous Work: Actions like research_start and studio_create run in the background. You must poll research_status or studio_status to determine when they are complete.
Follow-ups Are Mandatory: When using notebook_query, NotebookLM may provide truncated or narrowly-focused answers based only on the sources. Check if "Is that ALL you need to know?" applies, and proactively query again to fill gaps before answering the user.

Important Note

This skill replaces the legacy Python-based run.py wrapper. If you see old references to scripts/run.py or auth_manager.py in your history, ignore them. You are dealing with native MCP tools now.

Source

git clone https://github.com/pablodiegoo/Data-Pro-Skill/blob/main/.agent/skills/notebooklm/SKILL.mdView on GitHub

Overview

This skill lets you work directly with Google NotebookLM (notebooklm.google.com) using native MCP tools. It supports querying notebooks, searching documents, adding sources (URLs, Drive, PDFs), generating audio/video from documents, sharing notebooks, and managing your NotebookLM library—always grounding answers in your sources.

How This Skill Works

You verify your connection, identify your goal, and execute via native MCP tools. If authentication fails (401 Unauthorized or Invalid CSRF), prompt the user to run nlm login to re-authenticate. NotebookLM tasks can be asynchronous, so you may need to poll status endpoints; when using notebook_query, follow-up questions are often required to ensure a complete, grounded answer.

When to Use It

Query an existing notebook to retrieve a grounded summary or answer using NotebookLM sources.
Search documents within a notebook to locate specific information or citations.
Add sources to a notebook, including URLs, Google Drive files, or PDFs, to enrich future answers.
Generate an audio podcast or video overview from a document for quick consumption.
Share notebooks with teammates or manage notebook and library settings.

Quick Start

Step 1: Ensure your NotebookLM account is connected; if you get authentication errors, run nlm login in your terminal.
Step 2: Decide your goal (query notebook, add sources, or generate an artifact) and choose the appropriate MCP action (notebook_query, source_add, or studio_create).
Step 3: Execute the action via MCP tools and monitor status with the relevant status endpoint until the task completes.

Best Practices

Ask clarifying questions before running a notebook_query to ensure the returned information is complete and grounded.
Always require explicit confirmation (confirm=True) for destructive actions (notebook_delete, source_delete) and generative actions (studio_create, source_sync_drive).
If you encounter authentication errors, do not troubleshoot—prompt the user to run nlm login and re-authenticate.
For asynchronous tasks, poll progress with research_status or studio_status and report back when complete.
Cite NotebookLM sources in your answers and use the provided sources to avoid hallucinations.

Example Use Cases

A user asks for the latest policy summary; you query the notebook and return a grounded answer with citations to notebook sources.
A document is added as a PDF source to a notebook to improve future search results.
You generate an audio overview of a lengthy document and provide a playback link or embedded audio.
You share a notebook with a teammate, adjusting access permissions as needed.
You sync a Google Drive folder as a source to keep notebook content up to date.

Frequently Asked Questions

Add this skill to your agents