Audio Studio MCP
A local MCP server generates music and sound effects, builds loops, adjusts loudness and exports audio for games or video.
- 592 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- The server downloads model weights and writes or processes audio files in the chosen directory.
Install
- Works in
Manual install
git clone https://github.com/kanaworksai-hue/audio-studio-mcp.git
cd audio-studio-mcp
./install.shThe first README block targets macOS Apple Silicon. Setup creates an environment and downloads the model.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The server downloads model weights and writes or processes audio files in the chosen directory.
This is third-party code. Review the repository files before installing.
What it does
Audio Studio MCP exposes a local Stable Audio 3 model to an agent. Text prompts generate music or sound effects, and server tools help choose a candidate and prepare output. Users can create loops, fades, loudness adjustments and WAV, OGG or FLAC exports. Tools can edit or extend existing audio. A JSON sidecar retains the prompt, seed and settings. Apple Silicon uses MLX, while Windows with NVIDIA uses PyTorch. Generation runs on the user’s computer.
Who it is for. For game and video creators generating local background tracks and effects through MCP.
Good fit when
- Create a short music loop for a game.
- Generate a scene sound effect from a description.
- Normalize an existing clip’s loudness.
- Extend a track or revise a time range.
Not a fit when
- The machine lacks supported hardware.
- Model and runtime storage are unavailable.
- You expect an exact performance without listening to the output.
Example request
Create two short rain effects for a game and show their settings. After listening, prepare an OGG loop without an audible seam.MCP
- Transport
- stdio
- Authentication
- not required
Limitations
Local generation needs an Apple Silicon Mac or Windows with at least 8 GB NVIDIA VRAM, plus around 6 GB disk space on Mac or 14 GB on Windows. On Windows, sign into Hugging Face and accept the weight-access terms before downloading. Weights and generated audio are covered by the Stability AI Community License, separately from the server’s MIT code; check commercial-use terms. Before exporting, listen for loop seams, check loudness and compare the result with the brief.
How to disable. Remove the MCP connection and stop the server. Delete model weights, the environment and generated audio separately.
FAQ
Does generation need an API key?
No. Inference is local; Windows requires access to Hugging Face weights.
Which Python versions are supported?
Python 3.10 through 3.13.
Which export formats are available?
WAV, OGG and FLAC, with settings chosen for the editor or engine.
Is ffmpeg required?
It is listed among installation prerequisites.
Are generation settings retained?
Yes. A JSON sidecar stores prompt, seed and settings.
Discussion
No questions or reviews yet.
Related
MCP serversA local sound studio with an MCP server: an agent makes effects, music, songs and speech on open-weight models
MCP server for Audacity 3.x with 144 tools and 9 pipelines for audio cleanup, mastering and transcription, all processed locally.
MCP server for a Windows music studio: the agent writes lyrics, generates a song locally, arranges, mixes and exports WAV
An MCP server for Ableton Live: the agent creates tracks, places clips and builds arrangements straight in the open project
A local MCP server for voice work on macOS: voice cloning, speech synthesis and audiobooks from PDF on an agent's command
Hosted Magic Hour MCP server: the agent generates and edits video, images and audio through the service API
README in short
The server combines generation with finishing tools, letting an agent prepare editor-ready output after a candidate is selected.