Notebooklm Integration

The Problem

Google NotebookLM is a powerful research tool — you can add sources, generate AI podcasts, create video overviews, and chat with your documents. But every interaction requires the browser: navigate to the notebook, paste a URL, wait for processing, click generate, wait again, download the audio. If you're building a research pipeline or need to process 15 sources across 3 notebooks, the browser workflow becomes the bottleneck. This skill wraps the notebooklm-py Python SDK to give your coding agent full programmatic control: create notebooks, add sources (URLs, text, files), generate audio/video/slide artifacts, query notebooks via chat, and download outputs — all from the terminal.

What You Get

• Full notebook lifecycle — create, list, rename, and delete notebooks programmatically.
• Source management — add URLs, paste text, or upload files as notebook sources via CLI or Python SDK.
• Artifact generation — trigger audio (podcast), video overview, and slide deck generation with custom prompts (e.g., "make it engaging, 10-minute format").
• Chat with notebooks — ask questions about your sources and get grounded answers from NotebookLM's RAG system.
• Batch automation — process multiple notebooks in a loop: add sources, generate artifacts, download outputs, all without touching the browser.
• Proxy configuration — built-in handling for regions where NotebookLM blocks direct access (Hong Kong, China), with automatic SOCKS5 proxy routing.

Why This Beats Prompting It Yourself

You can't "prompt" NotebookLM to do things from the terminal — it's a browser-only product with no public API. This skill bridges that gap by wrapping the unofficial Python SDK and handling the messy parts: authentication via browser cookies, proxy routing for geo-blocked regions, artifact download with retry logic, and rate limit spacing. What takes 5 clicks and 3 minutes in the browser happens in one command.

Example Session

You: "Create a notebook with these 4 research papers and generate a 10-minute podcast overview."

Agent:

HTTPS_PROXY=socks5h://localhost:10808 notebooklm create "AI Chip Research"

HTTPS_PROXY=socks5h://localhost:10808 notebooklm source add "https://arxiv.org/abs/2601.12345" -n nb-abc123
HTTPS_PROXY=socks5h://localhost:10808 notebooklm source add "https://arxiv.org/abs/2602.67890" -n nb-abc123
HTTPS_PROXY=socks5h://localhost:10808 notebooklm source add "https://arxiv.org/abs/2603.11111" -n nb-abc123
HTTPS_PROXY=socks5h://localhost:10808 notebooklm source add "https://arxiv.org/abs/2604.22222" -n nb-abc123

HTTPS_PROXY=socks5h://localhost:10808 notebooklm generate audio "make it engaging, focus on practical applications, 10 minutes" -n nb-abc123 --wait

HTTPS_PROXY=socks5h://localhost:10808 notebooklm download audio ./ai-chip-podcast.mp3 -n nb-abc123

Use Cases

• Building a weekly research digest pipeline that automatically sources, summarizes, and generates audio from new papers in your field.
• Processing conference proceedings — batch-add 20+ papers and generate a podcast summary for your commute.
• Automating competitive intelligence — add competitor blog posts and documentation URLs, then chat with the notebook for quick analysis.
• Creating educational content — generate podcast-style explainers from technical documents for team onboarding.
• Research note organization — programmatically create topic-specific notebooks and populate them from RSS feeds or curated URLs.

Known Limitations

Requires notebooklm-py Python package (install via pip). NotebookLM blocks Hong Kong IP addresses — a SOCKS5 proxy is required for all operations except login. Authentication uses browser cookies, which expire periodically and need re-login. Artifact generation (especially audio) can take 2-5 minutes per notebook. Rate limits apply — space requests with 5-second delays (429s if faster). The SDK is unofficial and may break if Google changes NotebookLM's internal API.