NotebookLM RAG Agent

A full-stack web application that queries Google NotebookLM notebooks and generates professional research reports from conversation history. Combines Claude AI's agent capabilities with a React frontend to create an intelligent research assistant.

Features

• Multi-turn Research Conversations - Maintain context across multiple questions
• Source Citations - Track and display notebook sources in responses
• Document Generation - Create professional DOCX reports automatically
• Session Management - Save and retrieve conversation history
• Korean Font Support - Documents use "맑은 고딕" font

Tech Stack

Layer	Technology
Backend	Node.js + TypeScript + Express
Frontend	React 19 + TypeScript + Tailwind CSS + Vite
AI	Anthropic Claude Agent SDK
Documents	docx library
NotebookLM	Python + Patchright (browser automation)

Project Structure

notebook-lm-rag/
├── src/                    # Backend source
│   ├── api/                # Express route handlers
│   ├── mcp/tools/          # MCP tools (list, query, generate)
│   ├── session/            # Session management
│   ├── docx/               # Document generation
│   └── utils/              # Python runner, library reader
├── client/                 # React frontend
│   ├── src/components/     # UI components
│   ├── src/hooks/          # Custom React hooks
│   └── src/services/       # API client
├── .claude/skills/         # Claude Code skills
│   └── notebooklm-skill/   # NotebookLM Python automation
├── output/                 # Generated DOCX files
└── docs/                   # Documentation

API Endpoints

Method	Endpoint	Description
GET	/api/notebooks	List available notebooks
POST	/api/sessions	Create new research session
GET	/api/sessions/:id	Get session with Q&A history
POST	/api/sessions/:id/message	Send query to notebook
POST	/api/sessions/:id/generate-doc	Generate DOCX report
GET	/api/sessions/:id/document	Download generated document

Setup

Prerequisites

• Node.js 18+
• Python 3.10+ (for NotebookLM skill)
• Anthropic API key

Installation

# Install backend dependencies
npm install

# Install frontend dependencies
cd client && npm install

NotebookLM Skill Setup

cd .claude/skills/notebooklm-skill

# Create virtual environment
python -m venv .venv
.venv\Scripts\activate  # Windows
source .venv/bin/activate  # Linux/Mac

# Install dependencies
pip install -r requirements.txt

# Setup authentication (opens browser for Google login)
python scripts/auth_manager.py setup

Running

# Start backend (port 3000)
npm run dev

# Start frontend (port 5173)
cd client && npm run dev

Usage

1. Open http://localhost:5173
2. Select a notebook from the sidebar
3. Create a new research session
4. Ask questions - responses include NotebookLM citations
5. Click "Generate Document" to create a DOCX report

MCP Tools

The backend exposes three MCP tools for Claude Agent:

• list_notebooks - Enumerate available NotebookLM notebooks
• query_notebook - Ask questions to a specific notebook
• generate_document - Create DOCX research report from session Q&A

License

MIT