Video summary

Claude Code + NotebookLM = CHEAT CODE

Main summary

Key takeaways

Technology

Summary (Technological concept + workflow)

The video claims Claude Code can act as a powerful “research agent,” but the speaker argues it becomes dramatically better by adding NotebookLM as a secondary tool.

Instead of relying on Claude Code’s basic web-search workflow, the proposed pipeline is:

  1. Search YouTube (via a custom Claude Code skill) to find relevant/trending videos.
  2. Scrape/capture YouTube sources (using a Python-based scraper relying on yt-dlp to pull metadata like titles, views, authors, durations, and dates).
  3. Send the video URLs/captions to NotebookLM using a NotebookLM skill integrated into Claude Code.
  4. Use NotebookLM’s internal workflow (“scrape → RAG → analysis → deliverables”) to automatically produce:
    • Analysis outputs (e.g., top “Claude skills” and emerging trends)
    • Rich deliverables, including:
      • infographics
      • slide decks
      • podcasts/audio overviews
      • mind maps
      • flashcards

A key performance/cost claim is that most of the heavy work is offloaded to NotebookLM, so Claude Code spends only minimal tokens for orchestration—described as “virtually zero tokens” (i.e., no need to build/maintain a paid RAG stack).

Demo highlights / what the system produced

In a demo, the speaker says Claude Code:

  • Uploaded 20 YouTube sources into NotebookLM for analysis.
  • NotebookLM returned:
    • the top five Claude Code skills
    • emerging trends for how those skills are used
    • a handwritten blueprint-style infographic as a generated deliverable

The speaker emphasizes verification:

NotebookLM’s content is presented as grounded in the referenced videos/captions (not made up), and the sources/metadata are visible in both tools.

Setup / tutorial steps (how to connect the tools)

Unofficial NotebookLM connector (no public API)

The speaker notes that NotebookLM has no public API, so they use an unofficial connector:

  • NotebookLM-dash-py (GitHub repo), described as acting like an unofficial Python API.

YouTube search/sourcing skill for Claude Code

Provided via a custom skill that uses yt-dlp to scrape YouTube metadata. Two options:

  • Have Claude Code build the skill/script
  • Download an “MD file / setup” and provide it to Claude Code (via the creator’s community)

NotebookLM installation/authentication

  • Install by running terminal commands (copy/paste into a second terminal).
  • Run a NotebookLM login command that opens a browser window for authentication once.

Install/activate the NotebookLM skill inside Claude Code

After NotebookLM-side setup, the user installs/activates a Claude Code “skill” that teaches Claude Code how to:

  • create notebooks
  • generate deliverables via NotebookLM

Tool usage examples / commands

The video describes using the YouTube skill as either:

  • a slash command, or
  • plain-language prompts

Example parameterized query:

  • YT-search with inputs like:
    • query (e.g., “Claude Code skills”)
    • count (how many results/sources to fetch)

Key takeaways (why this matters)

The speaker argues the major value is not just automation, but that NotebookLM abstracts away a complex workflow that would otherwise be:

  • time-consuming
  • brittle (especially when building corpora from YouTube for agent workflows)
  • expensive if you replicate scraping/RAG/analysis/deliverable generation with another stack

Main speakers / sources

  • Speaker/creator: “Chase” (presenter; mentions “Chase AI Plus” and “Chase AI Plus community”)
  • External source referenced for integration: Tang Ling
  • GitHub repo referenced: NotebookLM-dash-py
  • Technology dependency referenced: yt-dlp
  • Target tools/platforms: Claude Code and NotebookLM

Original video