🎬 YTPipe - AI-Native YouTube Processing Pipeline

Transform YouTube videos into LLM-ready knowledge bases with a production-ready MCP backend.

Quick Start • Features • Documentation • MCP Tools

✨ Features

• 🤖 MCP Integration - 12 AI-callable tools for seamless agent integration

• 🎯 Smart Chunking - Semantic text chunking with timeline timestamps

• 🧠 Vector Embeddings - 384-dimensional embeddings for semantic search

• 🔍 Full-Text Search - Context-aware transcript search

• 📊 SEO Intelligence - AI-powered title, tag, and description optimization

• ⏱️ Timeline Analysis - Topic evolution and keyword density tracking

• 🏗️ Microservices - 11 independent, composable services

• 🔐 Type-Safe - Pydantic models throughout

• ⚡ Async-First - Non-blocking I/O operations

• 🗄️ Multi-Backend - ChromaDB, FAISS, Qdrant support

🚀 Quick Start

# Install
git clone https://github.com/leolech14/ytpipe.git
cd ytpipe
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt

# Process a video
ytpipe "https://youtube.com/watch?v=dQw4w9WgXcQ"

Result: Metadata + Transcript + Semantic Chunks + Embeddings + Vector Storage

🎯 Usage Examples

MCP Server (AI Agents)

python -m ytpipe.mcp.server

Then from Claude Code:

"Process this video: https://youtube.com/watch?v=VIDEO_ID"
"Search video dQw4w9WgXcQ for 'machine learning'"
"Optimize SEO for video dQw4w9WgXcQ"

CLI (Humans)

# Basic
ytpipe "https://youtube.com/watch?v=VIDEO_ID"

# Advanced
ytpipe URL --backend faiss --whisper-model large --verbose

Python API (Developers)

from ytpipe.core.pipeline import Pipeline

pipeline = Pipeline(output_dir="./output")
result = await pipeline.process(url)

print(f"✅ {result.metadata.title}")
print(f"   Chunks: {len(result.chunks)}")
print(f"   Time: {result.processing_time:.1f}s")

📋 MCP Tools

Pipeline (4 tools)

• ytpipe_process_video - Full pipeline

• ytpipe_download - Download only

• ytpipe_transcribe - Transcribe audio

• ytpipe_embed - Generate embeddings

Query (4 tools)

• ytpipe_search - Full-text search

• ytpipe_find_similar - Semantic search

• ytpipe_get_chunk - Get chunk by ID

• ytpipe_get_metadata - Get video info

Analytics (4 tools)

• ytpipe_seo_optimize - SEO recommendations

• ytpipe_quality_report - Quality metrics

• ytpipe_topic_timeline - Topic evolution

• ytpipe_benchmark - Performance analysis

🏗️ Architecture

MCP Server (12 tools) → Pipeline Orchestrator → 11 Services → Pydantic Models

Services:

• Extractors (2): Download, Transcriber

• Processors (4): Chunker, Embedder, VectorStore, Docling

• Intelligence (4): Search, SEO, Timeline, Analyzer

• Exporters (1): Dashboard

8 Processing Phases:

1. Download → 2. Transcription → 3. Chunking → 4. Embeddings →

2. Export → 6. Dashboard → 7. Docling → 8. Vector Storage

📊 Performance

🔧 Requirements

• Python 3.8+

• FFmpeg (for audio extraction)

• 4GB+ RAM recommended

• GPU optional (CUDA for acceleration)

📖 Documentation

• Quick Start

• Setup Guide

• Architecture

• API Reference

• Development Guide

🤝 Contributing

Contributions welcome! Please read CONTRIBUTING.md first.

📝 License

MIT License - see LICENSE for details.

🙏 Credits

Built with:

• FastMCP - MCP server framework

• OpenAI Whisper - Speech-to-text

• sentence-transformers - Text embeddings

• Model Context Protocol - AI tool standard

📧 Contact

Leonardo Lech

• Email: leonardo.lech@gmail.com

• GitHub: @leolech14

⭐ Star this repo if you find it useful!

Transform YouTube → Knowledge Base in seconds