whisper-mcp-server

FastMCP server for audio transcription using Whisper.

Installation

Via Smithery (Recommended)

npx -y @smithery/cli install @tapiocapioca/whisper-mcp-server --client claude

Via pip (Manual)

pip install whisper-mcp-server

Then configure MCP:

{
  "mcpServers": {
    "whisper": {
      "command": "python",
      "args": ["-m", "whisper_mcp"]
    }
  }
}

Prerequisites

• Python 3.10+
• Whisper server running on localhost:9102

Start the container:

cd brainery-containers
docker-compose up -d whisper-server

Tool

transcribe_audio

Transcribe audio file using Whisper.

Parameters:

• file_path (required): Absolute path to audio file
• language (optional): Language code or 'auto' for detection (default: auto)
• model (optional): Whisper model size - tiny/base/small/medium/large (default: base)

Supported formats: MP3, M4A, WAV, OGG, FLAC, WEBM

Returns:

{
  "text": "Transcribed text content...",
  "language": "en",
  "duration": 120.5,
  "segments": 15,
  "processing_time_ms": 3500
}

Usage Example

# Transcribe with auto language detection
result = transcribe_audio(
    file_path="/path/to/audio.mp3",
    language="auto",
    model="base"
)

# Transcribe Italian audio with larger model
result = transcribe_audio(
    file_path="/path/to/podcast.m4a",
    language="it",
    model="medium"
)

License

MIT