NotebookLM Audio Generator

Automates uploading multiple sources (files, URLs, YouTube, Drive, text) to a NotebookLM notebook, generating a deep dive audio overview in a preferred language, and downloading the result. It guides the user through notebook management, source addition, audio generation, and downloading using the notebooklm-mcp-cli.

NotebookLM Audio Generator Skill

This skill automates the process of using Google's NotebookLM to generate a deep-dive audio overview from various sources and download it to a specified directory.

Prerequisites

This skill relies on the notebooklm-mcp-cli package and the epub2txt utility for .epub support. Before proceeding, ensure the user has them installed and authenticated.

Installation & Setup

If the user has not set up the CLI or the conversion tool, instruct them to do so first:

1. Install the NotebookLM CLI:

   bash

   uv tool install notebooklm-mcp-cli
   # OR
   pip install notebooklm-mcp-cli
   

2. Authenticate:

   bash

   nlm login
   

3. Install epub2txt (for .epub support): The skill expects epub2txt to be installed in your programs directory. Clone and initialize it using the following commands:

   bash

   mkdir -p ~/Programs
   cd ~/Programs
   git clone https://github.com/SPACESODA/epub2txt.git
   cd epub2txt
   chmod +x run.sh
   ./run.sh  # This initializes the virtual environment
   

Workflow

When activated, follow these steps strictly in order. Do not skip steps.

Step 1: Gather Information

Ask the user for the following information one by one to prepare for the generation:

1. Ask for a desired name for the new notebook. If not provided, use the default name "Audio Overview Notebook".
2. Source Selection: Ask the user to select their sources from the following options:
   • Local Files: .pdf, .txt, and .epub (which will be automatically converted). Use GUI pickers:
     • macOS: osascript -e "set theFiles to choose file with prompt \"Select your source file(s):\" multiple selections allowed true" ...
     • Linux: zenity --file-selection --multiple ...
     • Windows: PowerShell OpenFileDialog.
   • Web/Video URLs: Ask the user to provide any website or YouTube URLs.
   • Google Drive: Ask for Google Drive Document IDs.
3. Preferred Language: Ask the user for the preferred output language (BCP-47 code).
   • Options: Provide common choices: en (English - default), zh (Chinese), ja (Japanese), es (Spanish), fr (French), de (German).
4. Download Destination: Use a GUI directory picker to select where the audio file should be saved.

Crucial: Ask the user interactively to confirm they have provided/selected all the sources they wish to include before proceeding.

Step 2: Create a New Notebook

Use the notebook_create tool to create a new notebook with the provided name. Keep track of the notebook_id.

Step 3: Upload Sources

• Loop through all gathered sources:
  • EPUB Files: Convert to .txt first: cd ~/Programs/epub2txt && ./run.sh "<path>". Use the new .txt path.
  • Other Files: Use source_add(source_type="file", file_path="...").
  • URLs/YouTube: Use source_add(source_type="url", url="...").
  • Drive: Use source_add(source_type="drive", document_id="...").
• Always set wait=true to ensure sources are processed.

Step 4: Generate Audio Overview

• Use the studio_create tool to start the audio generation.
  • Set notebook_id to the ID.
  • Set artifact_type to audio.
  • Set audio_format to deep_dive.
  • Set audio_length to long.
  • Language: Use the user's selected BCP-47 code (e.g., zh).
  • Custom Prompt: If the selected language is NOT English (en), you MUST provide the following focus_prompt to encourage a longer, more detailed output:

    "Please provide an extremely detailed deep dive. Analyze each source file thoroughly without omitting any details. The conversation should be as long as possible, aiming for over 40 minutes."

  • Set confirm to true.

Step 5: Monitor Generation Status

• Inform the user: Explicitly tell the user that generating a long, deep-dive audio overview can take 5 to 15 minutes.
• Use the studio_status tool with action: status in a polling loop (using run_shell_command with sleep 300 between checks) until the status becomes completed.
• Note the audio_url and artifact_id when finished. Crucial: If there are multiple audio artifacts returned in the status, always identify and note the latest one (the one with the most recent created_at timestamp).

Step 6: Download the Audio

• Using the artifact_id of the latest audio, try using the download_artifact tool first to save the audio to the destination path with an .mp3 extension.

Step 7: Final Verification

• Verify the downloaded file using file <output_path>.
• Inform the user that the process is complete.