Video summary

Claude Code + NotebookLM + Obsidian = GOD MODE

Main summary

Key takeaways

Technology

Summary of Video Subtitles (Tech Concepts / Workflow)

The video demonstrates a practical “research workflow” that combines Claude Code, NotebookLM, Obsidian, and Claude Code’s Skill Creator to automate end-to-end research and content deliverables.

The core idea is to turn multiple steps (source collection → analysis → deliverables → documentation) into one reusable “super skill”, then store results in an Obsidian vault so Claude Code can improve over time.

What the combined workflow does (“research on steroids”)

  1. Collect sources in Claude Code
    • Example source: YouTube videos.
  2. Use a specific Claude Code skill to fetch/structure video data
    • Example approach: use yt-dlp for YouTube search.
  3. Send that data to NotebookLM
    • NotebookLM is called from Claude Code through skills.
  4. NotebookLM performs analysis and can produce deliverables, such as:
    • podcast / video / infographic / slide deck (and others explicitly listed)
    • mind map, flash cards, audio review, infographic, etc.
  5. Return results to Claude Code, then store them as markdown/text files in an Obsidian vault.

The creator emphasizes flexibility: the “source” and even the NotebookLM part can be swapped for other inputs/flows as long as the template stays the same.

How the tooling is connected

1) Claude Code + Skill Creator

  • You create sub-skills using commands like /skill creator.
  • Example: build a YouTube search skill:
    • searches videos by query using yt-dlp
    • returns structured results
  • Then build a second layer: combine sub-skills into a single pipeline/super-skill that:
    • calls the YouTube search skill
    • sends results to NotebookLM
    • returns the requested deliverable (e.g., infographic)

2) NotebookLM integration (via GitHub repo / CLI)

  • NotebookLM has no public-facing API, so integration uses a GitHub tool (“notebook LM-py”).
  • Steps described:
    • install the repo via terminal commands
    • authenticate via a “space login” flow (browser opens for NotebookLM login)
    • use/install a NotebookLM skill (via a NotebookLM skill install), or instruct Skill Creator to create the NotebookLM skill from the repo

3) Obsidian as the persistent “memory” layer

  • Claude Code outputs analysis results into an Obsidian vault as markdown files.
  • Benefits highlighted:
    • humans can browse and connect notes (links/backlinks, graphs)
    • Claude Code can “see” the structure via transparent markdown files
    • enables a long-term improvement loop

The video describes a symbiotic/self-improving loop:

  • Run the workflow repeatedly
  • Store new analysis + deliverables in the vault
  • Over time, Claude Code learns/refines how it communicates and thinks
  • The claude.md file acts like a “brain within a brain,” capturing conventions and preferences
    • It can be updated based on new conversations to maintain output style over time.

Setup + Execution steps (tutorial-style)

  • Install Skill Creator in Claude Code:
    • use /plugin
    • search/install Skill Creator
    • restart Claude Code afterward
  • Create skills:
    • use /skill creator
    • describe desired behavior (example: a YouTube search skill that returns structured results)
    • optionally skip or run tests/evals
  • Create NotebookLM skill via GitHub integration tool:
    • install “notebook LM-py” with terminal commands
    • authenticate with NotebookLM
    • either install a NotebookLM skill directly or instruct Skill Creator to create it from the repo/commands
  • Combine into a pipeline super-skill:
    • use Skill Creator again to stitch YouTube → NotebookLM → deliverable into one skill
  • Run the pipeline:
    • example prompt: search YouTube for content about “MCP servers / Claude Code / MCP”
    • request:
      • top five MCP servers
      • analysis of what drives views, outliers, gaps, opportunities
    • request: generate an infographic

Performance / output timing

  • NotebookLM analysis for text: fairly quick (example: ~6 minutes)
  • Deliverables like slide decks: can take longer (example: up to ~15 minutes)
  • Infographics: take “a handful of minutes”

Example output described

  • The produced infographic (for “MCP”) includes items/companies such as:
    • Supabase
    • Context 7
    • Playwright
    • plus other named tooling mentioned in results (e.g., Figma, Sentry, PostHog)
  • The research also returns a full markdown file containing:
    • key takeaways
    • server entries
    • double-bracket links compatible with Obsidian navigation
    • backreferences and graph connectivity inside Obsidian

Main speakers/sources

  • Speaker/creator: The video narrator (speaker uses “I” throughout; no name provided in subtitles)
  • Key external sources/tools mentioned:
    • NotebookLM (integrated via GitHub repo “notebook LM-py”)
    • yt-dlp (YouTube searching)
    • Obsidian (vault/markdown storage)
    • Claude Code’s Skill Creator
  • Sponsor mentioned: “Yours truly” / promotion for Chase AI Plus and the Chase AI community (no separate human sponsor name given in subtitles)

Original video