kanaworksai-hue/audio-studio-mcp

Audio Studio MCP – Local AI Music & SFX Server

What it is – A small self‑hosted server that lets Claude Code, Claude Desktop, Cursor or any other MCP‑compatible client ask for background music, instrument loops or sound‑effects in plain English. The request is processed locally on your Mac (Apple Silicon) or Windows PC (NVIDIA GPU) by Stable Audio 3, an open‑source diffusion model for audio generation. The server returns ready‑to‑use WAV/OGG/FLAC files together with a JSON side‑car that records the prompt, seed and settings.

---

Key features (from the README)

Feature	What it does
Generate	Text‑to‑audio for music, instrument stems or SFX. 1‑4 candidates per call, optional BPM for music.
Edit	Re‑style a whole clip, in‑paint a time range, or continue (extend) a track.
Finish	Post‑process: seamless looping, fades, loudness normalisation (LUFS), conversion to 44.1 kHz 24‑bit WAV/OGG/FLAC.
Local	Runs entirely on‑device – Apple Silicon via MLX or NVIDIA GPU via PyTorch. No cloud, no API key, no per‑generation cost.
Reproducible	Each output gets a .json side‑car with the exact prompt, seed and parameters.
MCP integration	Exposes generate_audio, edit_audio, process_audio and unload_models tools that MCP clients can call.

---

Quick start (summary)

macOS (Apple Silicon)

git clone https://github.com/kanaworksai-hue/audio-studio-mcp.git
cd audio-studio-mcp
./install.sh          # sets up uv, creates .venv, clones Stable‑Audio‑3 runtime, downloads the medium model (~6 GB)

The installer prints the exact command you need to register the server with Claude:

claude mcp add --scope user audio-studio -- /path/to/audio-studio-mcp/.venv/bin/audio-studio-mcp

Windows (NVIDIA GPU)

git clone https://github.com/kanaworksai-hue/audio-studio-mcp.git
cd audio-studio-mcp
powershell -ExecutionPolicy Bypass -File install.ps1

You will be asked for a Hugging Face token the first time (the PyTorch weights are gated). After installation, register the server:

claude mcp add --scope user audio-studio -- "C:\path\to\audio-studio-mcp\.venv\Scripts\audio-studio-mcp.exe"

---

How to use (via an MCP client)

Add the server to the client’s MCP config, e.g.:

{
  "mcpServers": {
    "audio-studio": { "command": "/path/to/audio-studio-mcp/.venv/bin/audio-studio-mcp" }
  }
}

Then you can ask the model in natural language, for example:

• “Make a 60‑second lo‑fi BGM loop at 85 BPM for a cafe scene, give me 3 candidates.”
• “I need a heavy wooden door slamming in a stone hallway. Export it as OGG for Godot.”
• “Extend this track to 90 seconds and normalize it to –16 LUFS.” The client will call the generate_audio, edit_audio or process_audio tools exposed by the server and receive the processed audio file plus its JSON metadata.

---

Performance snapshot (from the README)

Platform	Task	Approx. time
M1 Mac (medium model)	20 s music (warm)	~10 s (≈2× real‑time)
RTX 3060 (medium model)	20 s music (warm)	~1.3 s
RTX 3060 (medium model)	Full 380 s track (warm)	~15 s
Medium model memory	Mac: 5‑7 GB RAM, Windows: ~5.3 GB VRAM

---

Configuration (environment variables)

• AUDIO_STUDIO_OUTPUT_DIR – where generated files are written (default repo/outputs).
• AUDIO_STUDIO_SA3_MODEL – medium (default), sm-music or sm-sfx for smaller, faster models.
• AUDIO_STUDIO_DEVICE – cuda (default on Windows if available) or cpu.
• AUDIO_STUDIO_IDLE_UNLOAD_MIN – minutes of inactivity before the model is freed (default 10).
• Other vars control cache size, backend selection, ffmpeg path, etc.

---

License & commercial use

• The code is released under the MIT license.
• The model weights are under the Stability AI Community License, which permits commercial use for entities earning less than US $1 M per year (check the license for details).

---

Who might find this useful?

• Game developers who need royalty‑free loops or SFX without leaving the IDE.
• Video creators looking for quick background music.
• Anyone experimenting with AI‑generated audio who wants a fully offline workflow.

---

Roadmap (as listed)

• ✅ Apple Silicon backend (MLX)
• ✅ Windows NVIDIA backend (PyTorch)
• ⬜ LoRA support (upstream runtime already has it)

---

Bottom line: Audio Studio MCP is a genuine, open‑source tool that brings local, zero‑cost AI audio generation into any MCP‑compatible client, with a focus on game‑dev and video‑production workflows.