youtube-mcp-server

Additional capabilities include intelligent caching and optimized performance configurations. The server implements a file-based caching system to avoid redundant processing of the same video URL, storing transcription results in a designated `TRANSCRIPTIONS_DIR` for quick retrieval on subsequent requests. Performance is further optimized through configurable concurrency settings, where the `MAX_WORKERS` parameter controls the number of parallel threads for transcribing audio segments, allowing users to balance speed against CPU and memory usage. The server also supports various Whisper model sizes, from `tiny` to `large` and `turbo`, letting users choose the appropriate trade-off between transcription accuracy and resource consumption. These capabilities ensure the server can be tuned for different deployment environments, from lightweight development setups to production systems requiring high throughput.

The product works overall through a technical architecture composed of several integrated services: the DownloadService, VADService, WhisperService, and CacheService. When a transcription request is received, the DownloadService fetches the audio stream using `yt-dlp`. The audio is then passed to the VADService, which uses the Silero model to detect voice activity and split the audio into segments with added padding to avoid cutting off words. These segments are distributed to the WhisperService for transcription, leveraging parallel processing as defined by the MAX_WORKERS setting. The entire pipeline operates in memory where possible, and results are managed by the CacheService to prevent duplicate work. The server communicates via Server-Sent Events (SSE) transport, typically running at `http://127.0.0.1:8000/sse`, and integrates with MCP clients through a standard configuration.

Benefits and measurable outcomes for users include significant reductions in development time and complexity for building video-aware AI features. Developers no longer need to integrate and maintain separate libraries for YouTube downloading, audio processing, and speech recognition. The in-memory pipeline and caching lead to faster transcription times, especially for repeated video analyses. The support for 99 languages and translation opens up global content analysis without requiring separate language-specific models. AI agents gain reliable, structured access to video content, enabling new capabilities like real-time video summarization, multi-language content monitoring, and automated generation of show notes or chapter markers from any YouTube video, all through a simple, standardized protocol interface.

Concrete use cases with specific workflow examples include an AI research assistant that, when given a YouTube URL of a lecture, uses the `get_video_info` tool to retrieve the title and description for context, then calls `transcribe_video` to obtain a full transcript for summarization and key point extraction. Another example is a content moderation system that transcribes live-streamed videos using the server to automatically flag spoken content against policy guidelines, leveraging the multilingual support to monitor channels in various languages. A third workflow involves an educational platform where an AI tutor fetches transcriptions of educational videos, translates them into a student's native language using the translation parameter, and then generates interactive quizzes based on the transcribed content, all automated through the MCP server's tools.

The target users are primarily developers building AI agents and applications that require interaction with YouTube content, including those working on research tools, content analysis platforms, accessibility services, and educational technology. The server integrates seamlessly with any MCP-compatible client, such as those used with Claude or other AI platforms that support the Model Context Protocol. Its tech stack is Python 3.10+, utilizing `yt-dlp`, OpenAI's Whisper, Silero VAD, and FastMCP for the server implementation. It requires `ffmpeg` for audio processing. The project is open-source under the MIT license, with no explicit pricing plans mentioned, indicating it is freely available for use, modification, and distribution, with community contributions encouraged through the outlined fork and pull request workflow on GitHub.

In summary, the YouTube MCP Server provides a powerful, unified, and efficient bridge between AI agents and YouTube's video content, solving the key problems of metadata access and accurate transcription through a standardized protocol. By combining robust extraction, intelligent in-memory processing, extensive language support, and configurable performance, it enables developers to build sophisticated video-processing capabilities into their AI applications with minimal overhead. The primary value is delivering structured, actionable text and data from video, transforming opaque audio-visual content into a searchable, analyzable resource for the next generation of AI-powered tools.