🎙️ MCP Whisper

MCP (Model Context Protocol) server for audio transcription with multiple providers:

• 🌐 OpenAI API - Cloud-based, accurate
• 🖥️ whisper-cli - Local Python version
• ⚡ whisper.cpp - Fast C++ implementation

Installation

Global install (recommended)

npm install -g @clawbow/mcp-whisper

Local install

npm install @clawbow/mcp-whisper
npx mcp-whisper

Quick Setup

# Check dependencies and setup
npx mcp-whisper-setup

# Or auto-install whisper-cli
npx mcp-whisper-setup --auto

Configuration

Claude Desktop

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "whisper": {
      "command": "npx",
      "args": ["-y", "@clawbow/mcp-whisper"],
      "env": {
        "OPENAI_API_KEY": "sk-...",
        "WHISPER_PROVIDER": "auto"
      }
    }
  }
}

Environment Variables

| Variable | Description | Default | |----------|-------------|---------| | OPENAI_API_KEY | OpenAI API key | - | | WHISPER_PROVIDER | Provider: openai, local, whisper-cli, auto | auto | | WHISPER_LOCAL_URL | URL for whisper.cpp server | http://localhost:8080/inference | | WHISPER_CLI_PATH | Path to whisper binary | whisper |

Providers

1. OpenAI API (easiest)

export OPENAI_API_KEY="sk-..."
export WHISPER_PROVIDER="openai"
npx @clawbow/mcp-whisper

2. whisper-cli (local Python)

Install:

pip install -U openai-whisper
# Also needs ffmpeg
brew install ffmpeg  # Mac
sudo apt install ffmpeg  # Ubuntu

3. whisper.cpp (fastest local)

git clone https://github.com/ggerganov/whisper.cpp.git
cd whisper.cpp
make
./models/download-ggml-model.sh base
./server -m models/ggml-base.bin --port 8080

Available Tools

transcribe_audio

{
  "audio_path": "/path/to/audio.wav",
  "language": "fr",
  "provider": "auto"
}

transcribe_audio_base64

{
  "audio_base64": "base64encoded...",
  "filename": "audio.wav",
  "language": "fr",
  "provider": "auto"
}

list_providers

List available providers and their status.

Auto-Detection

When provider: "auto", the server tries in order:

1. OpenAI API (if OPENAI_API_KEY is set)
2. whisper-cli (if whisper is in PATH)
3. whisper.cpp local server (if responding on port 8080)

Usage Example

In Claude Desktop, you can say:

"Transcribe this audio file: /path/to/meeting.wav"

Claude will automatically use the available whisper provider.

License

MIT © DarkBow