# Audio Studio MCP

> Local MCP server that generates music and sound effects with Stable Audio 3, then edits and exports them for games and video.

Audio Studio MCP is an open-source MCP server from KanaWorks_AI that gives Claude Code, Claude Desktop, Cursor and other MCP clients a sound studio running on the user's own Mac or Windows PC. It generates background music, instrument loops and sound effects locally with Stable Audio 3. The README states there is no cloud service and no API key.

## What It Is

Audio Studio MCP exposes audio generation, editing and post-processing as MCP tools. A user asks their MCP client in plain language for a BGM track or a sound effect, and the server builds a prompt in the Stable Audio 3 format, runs the model locally, and writes the result to disk. Each output file has a `.json` sidecar recording its prompt, seed and settings, so the same prompt and seed can reproduce the same file.

## Tools and Workflow

- `generate_audio` turns text into music, instrument loops or stems, or sound effects. It returns 1–4 candidates per call and accepts a BPM value and structured SFX fields for source, action, material and space.
- `edit_audio` offers variation (re-style a whole clip), inpaint (regenerate a time range) and continue (extend a track).
- `process_audio` chains trimming, silence trimming, mono conversion, resampling, loudness normalization, seamless loop crossfade, fades and peak normalization, then exports WAV, OGG or FLAC without modifying the input.
- `unload_models` frees model memory immediately.

Generated files are 44.1 kHz 24-bit stereo WAV written to an outputs folder.

## Local Backends and Setup

On Apple Silicon the server runs Stable Audio 3 through MLX. On Windows it uses PyTorch with an NVIDIA GPU, with 8 GB of VRAM or more required. The macOS installer needs about 6 GB of disk and recommends 16 GB of RAM. The Windows install needs about 14 GB of disk and a Hugging Face account to accept the model license. Both installers use `git` and `ffmpeg`, and register the server with `claude mcp add`.

Linux with an NVIDIA GPU is described as untested and without an installer. Intel Macs and AMD GPUs are not supported. The model produces instrumental music and sound effects, not intelligible vocals.

## Licensing Notes

The project code is MIT-licensed. The Stable Audio 3 model weights fall under the Stability AI Community License, which governs use of the model and its output, so commercial users should read it first. The project says it is independent and not affiliated with Stability AI, Apple, NVIDIA or Anthropic.

## Features
- Text-to-music, instrument loop and sound effect generation
- 1–4 candidates per call with reproducible seeds
- Structured SFX prompts (source, action, material, space)
- Edit modes: variation, inpaint, continue
- Seamless loop crossfade, fades and loudness (LUFS) normalization
- Export to WAV, OGG and FLAC
- JSON sidecar with prompt, seed and settings
- Runs locally with no cloud or API key
- Optional local HTTP server shared between sessions
- Idle model unloading and unload_models tool

## Integrations
Claude Code, Claude Desktop, Cursor, Stable Audio 3, MLX, PyTorch, Hugging Face, ffmpeg

## Platforms
WINDOWS, MACOS, LINUX, API, CLI

## Pricing
Open Source

## Version
v0.2.1

## Links
- Website: https://kanaworksai-hue.github.io/audio-studio-mcp/
- Documentation: https://github.com/kanaworksai-hue/audio-studio-mcp#readme
- Repository: https://github.com/kanaworksai-hue/audio-studio-mcp
- EveryDev.ai: https://www.everydev.ai/tools/audio-studio-mcp