youtube-mcp-server

The second major feature group is Smart Transcription, which goes far beyond simple speech-to-text. The server implements an in-memory processing pipeline that downloads the audio stream, loads it directly into RAM, and then performs Voice Activity Detection (VAD) using Silero VAD for precise segmentation. This segmentation ensures that only actual speech is processed, reducing noise and improving accuracy. After segmentation, the audio chunks are transcribed using OpenAI's Whisper model, which supports an impressive 99 languages. Users can choose from multiple Whisper model sizes—tiny, base, small, medium, large, or turbo—to balance speed and accuracy. Additionally, the server includes a translation feature: you can request that the transcription be generated in any supported language, effectively translating the spoken content. This multilingual capability is particularly valuable for global teams analyzing videos in multiple languages or building AI assistants that communicate in the user's native tongue.

The third feature group encompasses Caching and Performance Optimizations that make the server suitable for production use. The server employs intelligent file-based caching, storing completed transcriptions in a configurable directory. This means repeated requests for the same video are served from disk without re-downloading or re-processing, drastically reducing latency and computational overhead. Furthermore, the server leverages hardware acceleration via MPS (Apple Silicon) or CUDA (NVIDIA GPUs) for Whisper inference, significantly speeding up transcription times. Parallel processing is implemented at the segment level: the MAX_WORKERS configuration parameter allows users to set the number of concurrent threads that transcribe audio segments simultaneously. This concurrency model maximizes throughput on multi-core machines. Together, these optimizations ensure that even long videos can be transcribed in a fraction of the time it would take with naive sequential processing, making the server practical for batch processing and real-time applications.

How the product works overall is straightforward. The server runs as a Python application using the SSE (Server-Sent Events) transport protocol, exposing an HTTP endpoint at http://127.0.0.1:8000/sse. AI agents connect to this endpoint via a standard MCP client configuration, which involves adding the server URL to the MCP client's settings. Once connected, agents can call two primary tools: get_video_info and transcribe_video. The first tool requires only a YouTube URL and returns a JSON object with all available metadata. The second tool accepts a URL and an optional language parameter—if set to 'auto', the server detects the video's spoken language and transcribes it; if a specific language code is provided, it transcribes and translates the content accordingly. The server handles all the underlying complexity: downloading the audio, applying VAD, selecting the Whisper model, managing caching, and returning timestamped transcription segments. This workflow is designed to be invisible to the calling AI agent, which simply receives structured data ready for further processing.

Concrete use cases for the YouTube MCP Server are diverse and practical. A content moderation AI agent can use get_video_info to quickly fetch metadata for a large playlist, flagging videos with inappropriate titles or descriptions. An educational platform's assistant can transcribe a lecture video and then translate it into the student's preferred language, making learning accessible across linguistic barriers. Researchers analyzing social media trends can automatically transcribe thousands of YouTube videos to extract spoken keywords and sentiments. For customer support bots, the server enables answering questions directly based on video tutorials—the bot transcribes the relevant video and then searches the text for answers. In all these scenarios, the outcome is the same: AI agents gain the ability to 'read' and 'understand' video content as easily as text, unlocking new automation possibilities and reducing manual effort. The timestamped output also allows agents to reference specific moments in a video, enabling rich interactions like jumping to a particular point.

The target audience for this server includes developers building AI assistants, data scientists processing video corpora, content analysts, and anyone who needs to integrate YouTube data into automated workflows. The server is written in Python and requires Python 3.10 or later, along with ffmpeg for audio processing. It runs on macOS, Linux, and Windows, with installation via git clone and the 'uv sync' command. The configuration is highly customizable through environment variables and the config.py file, allowing adjustments to Whisper model size, VAD parameters, audio sampling rate, segmentation padding, and concurrency level. The project is open source under the MIT License, making it free to use, modify, and distribute. Whether you are building a simple chatbot that answers questions about videos or a complex data ingestion pipeline, the YouTube MCP Server provides a reliable, performant foundation. Its primary value lies in giving AI agents streamlined access to YouTube's rich content, turning video into actionable structured data.