What it does
Audio Studio MCP lets Claude Code, Claude Desktop, Cursor and other MCP clients generate game audio locally. You describe the background music or sound effect you want, and the server runs Stability AI's Stable Audio 3 model on an Apple Silicon Mac (through MLX) or a Windows PC with an NVIDIA GPU (through PyTorch). Nothing is uploaded and there's no per-generation fee.
It has four tools. One generates music, instrument loops or sound effects with 1 to 4 candidates per call, a tempo setting and structured fields for an effect's source, action, material and space. One edits audio by restyling a clip, regenerating a time range or extending a track. One processes audio: trimming, resampling, seamless loops, fades, loudness or peak normalization, and export to WAV, OGG or FLAC. The last one frees model memory. Every file gets a JSON sidecar with its prompt, seed and settings so you can reproduce it.
The same server also has a local web page for generating, comparing, editing and browsing clips by hand, and the server passes Stable Audio 3's prompt guide to the agent. Setup runs an install script that downloads the model weights, about 6 GB on a Mac and about 14 GB on Windows including PyTorch. The model can't make vocals, and its weights are under the Stability AI Community License, which sets the terms for commercial use.
Features
- Generates music, instrument loops and sound effects locally, with no API key
- 1 to 4 candidates per request, with structured prompts for sound effects
- Restyle, regenerate part of, or extend a clip
- Seamless loops, fades, loudness normalization and WAV, OGG or FLAC export
- JSON sidecar with prompt, seed and settings for every file
Limits
- Needs an Apple Silicon Mac or a Windows PC with an NVIDIA GPU (8 GB+ VRAM); Linux is untested
- Large model download, and Windows needs a Hugging Face account to accept the model license
- No vocals; commercial use follows the Stability AI Community License