youtube-mcp-server

Product Introduction

1. Definition: youtube-mcp-server is a specialized Model Context Protocol (MCP) server designed for AI-driven extraction of YouTube video metadata and transcriptions. It operates as a middleware layer between AI agents and YouTube content, leveraging advanced audio processing and NLP technologies.
2. Core Value Proposition: It enables real-time, multilingual transcription and metadata retrieval without video downloads, solving data accessibility challenges for AI workflows. Primary keywords: YouTube transcription API, metadata extraction server, MCP protocol for video analysis.

Main Features

1. Metadata Extraction Engine: Uses yt-dlp to fetch video metadata (title, views, duration, tags, etc.) via direct API calls. Returns structured JSON with zero video downloads, reducing bandwidth by 95% compared to traditional scrapers.
2. In-Memory Transcription Pipeline:
   • How it works: Audio streams are processed in RAM (no disk I/O) → segmented via Silero VAD (Voice Activity Detection) → transcribed using OpenAI Whisper.
   • Tech stack: Whisper models (tiny to turbo) with CUDA/MPS acceleration, 99-language support, configurable SAMPLING_RATE (16kHz default).
3. Multilingual Translation: Translates transcriptions to any supported language (e.g., Japanese → English) via Whisper’s cross-lingual capabilities. Uses dynamic language codes (e.g., "fr" for French).
4. Intelligent Caching: File-based caching (transcriptions/ directory) stores processed data using video ID + language keys. Reduces redundant API calls and compute costs by 70% for repeat requests.
5. Parallel Processing: Concurrent segment transcription via thread pools (MAX_WORKERS=4 default). Scales linearly with CPU cores, cutting 30-minute video processing to <5 minutes.

Problems Solved

1. Pain Point: Manual transcription tools (e.g., Otter.ai) lack YouTube integration and require uploads. Keywords: slow video transcription, no native YouTube metadata API.
2. Target Audience:
   • AI Agent Developers: Building YouTube-summarizing agents or content analyzers.
   • Data Engineers: Needing structured video data for NLP pipelines.
   • Accessibility Teams: Auto-generating subtitles for multilingual content.
3. Use Cases:
   • Real-time video content moderation via transcript analysis.
   • Training LLMs on YouTube educational content with translated transcripts.
   • SEO analysis of video metadata at scale.

Unique Advantages

1. Differentiation vs. Competitors: Unlike Pytube (metadata-only) or Whisper Web UIs, it combines metadata + transcription in one MCP-standardized endpoint. Outperforms Google Speech-to-Text in cost (free/local) and language coverage (99 vs. 50+ languages).
2. Key Innovation: Silero VAD + Whisper in-memory pipeline with segment padding (SEGMENT_PADDING_MS=200). Eliminates disk I/O bottlenecks and improves word-boundary accuracy by 40% vs. standalone Whisper.

Frequently Asked Questions (FAQ)

1. Does youtube-mcp-server download YouTube videos?
   No. It extracts metadata via yt-dlp APIs and processes audio streams in-memory without video downloads.
2. What hardware is needed for GPU acceleration?
   Requires NVIDIA GPU (CUDA) or Apple Silicon (MPS) for Whisper models "medium" or larger. "Tiny" model runs on CPU-only systems.
3. How to handle long videos (>1 hour)?
   Increase MAX_WORKERS (e.g., 8) and use Whisper "large" model. Caching prevents reprocessing.
4. Is YouTube API key required?
   No. It uses public yt-dlp endpoints, avoiding YouTube Data API quotas.
5. Can it transcribe live streams?
   Yes, if the stream is archived on YouTube. Real-time live transcription is unsupported.