AudioGen MCP Server

# AudioGen MCP Server

## Project Overview

This is an MCP (Model Context Protocol) server that wraps Meta's AudioGen model for generating sound effects from text descriptions. It's designed to be used with Claude Code to generate game sound effects on-demand.

## Quick Start

```bash
# Install and run
uvx audiogen-mcp

# Or for development:
pip install -e .
audiogen-mcp
```

## Project Structure

```
audiogen-mcp/
├── src/audiogen_mcp/
│   ├── __init__.py
│   └── server.py      # Main MCP server implementation
├── pyproject.toml     # Package configuration
├── README.md          # Full documentation
└── CLAUDE.md          # This file
```

## Available MCP Tools

1. **generate_sound_effect** - Start a background generation job (returns job_id)
2. **check_generation_status** - Poll for job completion (use after generate_sound_effect)
3. **list_generation_jobs** - List all jobs and their status
4. **list_generated_sounds** - List previously generated files
5. **get_model_status** - Check model and device status

## Key Technical Details

- Uses `facebook/audiogen-medium` model
- Apple Silicon: Uses MPS (Metal) for GPU acceleration
- Generation speed: ~60s per 5s of audio on M-series chips
- Output: WAV files at model's native sample rate (16kHz)
- Default output directory: `~/audiogen_outputs/`

## Example Usage in Claude Code

Once configured as an MCP server, you can ask Claude Code to:

- "Generate an explosion sound effect for the game"
- "Create a batch of UI sounds: click, hover, and error"
- "Make a retro 8-bit power-up sound, 2 seconds long"

## Dependencies

- Python 3.9-3.11
- PyTorch with MPS support
- audiocraft (Meta's AudioGen)
- mcp (Model Context Protocol SDK)
- ffmpeg (system): `brew install ffmpeg`