MCP Video & Audio Text Extraction Server

The MCP Video & Audio Text Extraction Server provides standardized access to speech-to-text capabilities by downloading and transcribing audio from videos or audio files across multiple platforms. Built with MCP, it exposes clear tools for downloading videos, extracting audio, and transcribing both video and audio inputs, making it easy to integrate transcription workflows with large language models and other MCP-enabled clients.

How to use

You run the MCP server locally or in your environment and connect to it from an MCP client. The server offers four primary tools: download videos from supported platforms, extract audio from videos, transcribe text from videos, and transcribe text from audio files. Use these tools to build workflows that fetch media, convert it to audio when needed, and obtain accurate transcriptions in multiple languages. Start tasks asynchronously to handle long videos and large audio files efficiently.

To access the server from your MCP client, configure your client to point to the local or remote MCP endpoint and call the desired tool by its name. You can combine download and transcription steps as part of a single workflow, or run them separately depending on your use case.

How to install

Prerequisites you need before starting

• Python 3.10+ installed on your system
• FFmpeg installed for audio processing
• A runtime or package runner capable of executing MCP endpoints (see below for a recommended runner)

Install the runner and set up the MCP server with the recommended approach
- Use uv to directly run the video extraction server from the provided command
- The installation uses a bootstrap script to install uvx and prepare the environment

curl -LsSf https://astral.sh/uv/install.sh | sh

Install FFmpeg on your platform to enable audio processing. Choose the command that matches your OS and run it to install FFmpeg.

# Ubuntu or Debian
sudo apt update && sudo apt install ffmpeg

# Arch Linux
sudo pacman -S ffmpeg

# MacOS
brew install ffmpeg

# Windows (Chocolatey)
choco install ffmpeg

# Windows (Scoop)
scoop install ffmpeg

Configure the client to connect to the MCP server. The following example shows how to register the server in an MCP-enabled client that uses Claude/Cursor as an example:

{
  "mcpServers": {
    "video_extraction": {
      "command": "uvx",
      "args": ["mcp-video-extraction"]
    }
  }
}

Additional configuration and notes

Environment variables and tuning options let you tailor transcription quality and performance. The server supports configuring the Whisper model, language, download formats, and temporary storage. Use the following settings as a starting point and adjust to your workload.

Configuration details

Whisper settings and formats affect transcription results and performance. These options control the model size, language handling, and output format. You can also set how YouTube videos and audio extraction are performed and where temporary files are stored.

Example configuration (high level): It defines model size, language, download formats, and temp storage. Copy this into your deployment environment as a starting point.

{
  "WHISPER_MODEL": "base",
  "WHISPER_LANGUAGE": "auto",
  "YOUTUBE_FORMAT": "bestaudio",
  "AUDIO_FORMAT": "mp3",
  "AUDIO_QUALITY": "192",
  "TEMP_DIR": "/tmp/mcp-video",
  "DOWNLOAD_RETRIES": 10,
  "FRAGMENT_RETRIES": 10,
  "SOCKET_TIMEOUT": 30
}

Notes

The Whisper model file is downloaded on the first run and cached for subsequent uses. This may take several minutes depending on your network. Ensure you have enough disk space for temporary audio files and model downloads. A GPU is recommended for faster transcription, especially for long videos.

Security and best practices

Operate the server in a secure environment and restrict access to MCP clients you trust. Regularly update dependencies and monitor resource usage to prevent long-running tasks from exhausting system resources.

Troubleshooting

If you encounter slow downloads or transcription delays, verify network connectivity, ensure FFmpeg is accessible in your PATH, and confirm that the Whisper model has completed downloading on first run. Check that the chosen model size matches available system resources.

Available tools

Video download

Download videos from supported platforms using the integrated downloader (yt-dlp) to obtain the media content for processing.

Audio download

Extract the audio track from downloaded videos to prepare for transcription.

Video text extraction

Download the video and transcribe its spoken content into text.

Audio file text extraction

Transcribe text directly from audio files.