Type a topic, and AI finds YouTube videos, analyzes them, and creates podcasts, slides, and reports for you

한국어 | English

---

NLM Research

Type a topic you're curious about, and this tool finds related YouTube videos, has AI analyze them, and turns them into various content formats — automatically.

What You Can Create

Content	Description	Perfect for...
🎧 AI Podcast	Two AI hosts discuss the topic in a natural audio conversation (.mp3)	Learning on your commute — just listen and absorb
📊 Briefing Report	A clean, structured summary document (.md)	5-minute read before a meeting
🎬 Presentation Slides	Auto-generated slide deck (.pptx)	Cutting your presentation prep time dramatically
🧠 Mind Map	Visual map showing how topics connect	Seeing the big picture of complex subjects fast
💬 AI Q&A	Ask questions about your collected sources and get instant answers	When you don't have time to watch hours of video
🌐 Web Research	AI automatically finds more related sources from the web	When you need broader perspectives

All of this happens with a single command. Just type a topic, and the system finds YouTube videos, feeds them to AI, generates your chosen content, and saves everything to your computer.

How to Get Started

Step 1: Install the tools

Tool	Install Command	Purpose
Claude Code	Official install guide	AI assistant that runs everything
uv	curl -LsSf https://astral.sh/uv/install.sh | sh	Fast Python package manager
nlm	uv tool install notebooklm-mcp-cli	NotebookLM CLI + MCP server
yt-dlp	uv tool install yt-dlp	YouTube search

Verify your installation:

claude --version
nlm --version
yt-dlp --version

Alternative: install with pip (if uv is not available)

pip install notebooklm-mcp-cli yt-dlp

Step 2: Connect NotebookLM to Claude Code

The NotebookLM MCP server enables AI podcast, slides, and report generation. Without it, you'll see studio_create tool not found errors.

Option A: Automatic (recommended) — just run Claude Code inside this project directory:

cd nlm-research
claude

The .claude/settings.json in this repo auto-configures the MCP connection. No extra setup needed.

Option B: Manual — add to your global Claude Code config:

claude mcp add notebooklm-mcp -- notebooklm-mcp

Step 3: Connect your Google account

NotebookLM needs Google sign-in. Run this once and follow the browser prompt:

nlm login

Step 4: Run your first research

cd nlm-research   # Navigate to this folder
claude            # Start the AI assistant

# Now type this:
/research run AI agent trends --auto

Add --auto to run everything automatically without pausing. Remove it if you want to approve each step.

Note: The /research command is a Claude Code skill loaded from this project directory. Make sure you run claude from inside the nlm-research folder.

Use Case Guide

Choose the scenario that fits your needs.

"I have a presentation tomorrow and no slides"

/research run 2026 AI market outlook --preset presentation --auto

You get a 📊 briefing report + 🎬 slide deck (.pptx) automatically.

"I want to learn a new technology quickly"

/research run React 19 new features --preset learning --auto

You get a 📊 study guide + 🎧 AI podcast (mp3) + 📝 quiz.

"I need to understand market trends"

/research run AI agent 2026 trends --preset trend-report --auto

You get a 📊 trend analysis briefing report.

"I need a competitor analysis"

/research run competitor-X product review --preset competitor --auto

You get a 📊 SWOT analysis report.

"I want to go deep on a topic"

/research run LLM architecture comparison --preset deep-dive --auto

AI finds additional web sources and creates a 📊 deep analysis report.

"Just a quick research summary"

/research run AI agents --auto

Default preset: 📊 briefing report + 💬 Q&A analysis.

How It Works

Think of it like a research assistant: finds relevant books (search) → borrows them from the library (collect) → reads and summarizes them (analyze) → prints the report (export) — all done by AI.

Step	What happens	Think of it as...
🔍 Search	Finds related YouTube videos for your topic	Finding relevant books at the library
📚 Collect	Adds those videos to NotebookLM	Borrowing the books
🧠 Analyze	AI analyzes everything and creates your outputs	Reading and summarizing
📤 Export	Saves the finished files to your computer	Printing the report

Files You'll Get

Everything is automatically saved by topic under ~/research-output/<topic>/.

~/research-output/
├── AI_agent_trends/
│   ├── AI_agent_trends_report.md          # Briefing report
│   ├── AI_agent_trends_analysis.md        # Q&A deep analysis
│   ├── AI_agent_trends_podcast.mp3        # AI podcast audio
│   ├── AI_agent_trends_quiz.json          # Learning quiz
│   └── AI_agent_trends_slides.pptx        # Presentation slides
├── last_session.json                       # Last session reference
└── research_sessions.jsonl                 # Session history

File	What it is	Which presets create it
*_report.md	A structured briefing document synthesized by AI	All presets
*_analysis.md	Deep insights from AI Q&A sessions	All presets
*_podcast.mp3	An audio conversation between two AI hosts	learning
*_quiz.json	A quiz to test your understanding	learning
*_slides.pptx	A ready-to-present slide deck	presentation

Fine-tuning Your Research

Search options

Option	What it does	Default
-n <number>	How many YouTube results to fetch	10
-d	Sort by newest first	off

Run options

Option	What it does	Default
--auto	Run without asking for confirmation	off
--preset <name>	Choose a preset (see Use Case Guide above)	default
--top <N>	How many top videos to actually use	depends on preset
--notebook <id>	Add to an existing notebook instead of creating a new one	create new
--lang <code>	Language for generated content (e.g., en, ko)	ko

What's the difference between -n and --top?

• -n controls how many videos to search for on YouTube
• --top controls how many of those videos to actually use

# Search 15 videos, but only use the top 3
/research run AI -n 15 --top 3 --auto

Step-by-step manual execution

Instead of running everything automatically, you can run each step yourself:

/research search AI agents              # 1. Search YouTube
/research collect                        # 2. Collect selected videos into NotebookLM
/research analyze <notebook-id>          # 3. AI analysis
/research export <notebook-id>           # 4. Export files
/research status                         # Check current progress

If Something Goes Wrong

The system handles problems at three levels:

• 🟢 Most issues fix themselves. Things like expired login sessions are refreshed automatically. Rate limits are waited out and retried.
• 🟡 Partial failures are skipped gracefully. If one video can't be added (maybe it's private or deleted), the system skips it and continues with the rest.
• 🔴 Serious problems stop the process and tell you what to do. For example, if your login is completely broken or no videos were found at all.

Common fixes

What you see	What to do
deno install fails / JSR package not found	The old Deno/JSR installation method has been deprecated. Install via uv tool install notebooklm-mcp-cli instead (see Step 1)
studio_create tool not found / download_artifact not available	NotebookLM MCP server is not connected to Claude Code. Run claude inside the nlm-research directory (auto-configured), or run claude mcp add notebooklm-mcp -- notebooklm-mcp
/research command not found	Claude Code doesn't see the skill files. Make sure you cd nlm-research first, then run claude. Skills are loaded from the project directory
"NotebookLM authentication required"	Run nlm login in your terminal
Login keeps failing	Try nlm login switch default or clear your browser cache
A specific video fails to load	It's probably private or deleted — the system skips it automatically
Report generation fails	You may have too few or too many sources — check with /research status
0 sources collected	Try different search keywords

Project Structure

nlm-research/
├── CLAUDE.md                               # Project instructions for Claude Code
├── .claude/
│   ├── settings.json                       # MCP server + permissions config
│   ├── rules/
│   │   └── research-pipeline.md            # Pipeline rules
│   ├── skills/research/                    # Skill files (recommended)
│   │   ├── SKILL.md                        # Command router
│   │   ├── run.md                          # Full automated flow
│   │   ├── search.md                       # YouTube search
│   │   ├── collect.md                      # Source collection
│   │   ├── analyze.md                      # AI analysis
│   │   ├── export.md                       # File export
│   │   ├── status.md                       # Session status
│   │   ├── scripts/
│   │   │   └── youtube_search.py           # YouTube search script
│   │   └── references/
│   │       ├── nlm-commands.md             # NotebookLM reference
│   │       └── workflow-examples.md        # Workflow examples
│   └── commands/research/                  # Legacy (backward compatible)
├── README.md                               # This file (English)
├── README_KO.md                            # Documentation (Korean)
└── assets/                                 # Images used in documentation

Contributing

Contributions are welcome! Here's how:

1. Fork this repository
2. Create a feature branch (git checkout -b feat/amazing-feature)
3. Commit your changes (git commit -m 'feat: add amazing feature')
4. Push to the branch (git push origin feat/amazing-feature)
5. Open a Pull Request

License

This project is licensed under the MIT License.