Audio Studio MCP Runs Stable Audio 3 Locally on Mac and Windows Without the Cloud

⚡ The gist in 5 seconds - The kanaworksai-hue/audio-studio-mcp repository adds local generation of background music and sound effects using the Stable Audio 3 model. - On Apple Silicon it runs via MLX, on Windows machines with NVIDIA GPUs — via PyTorch; no cloud or API key required. - It connects to Claude Code and any other MCP client as a regular tool server. ### 🔍 What was found The audio-studio-mcp project is positioned as an MCP server for generating BGM and SFX directly on local hardware. The repository description explicitly lists two backends: MLX for Apple Silicon and PyTorch for Windows with NVIDIA graphics cards. The model is Stable Audio 3. The key feature is stated in the very first line: no cloud, no API key — meaning all inference runs on the user's machine, without contacting external services and without any credentials. The workflow is standard for MCP: the client (Claude Code or any other compatible one) sees the server as a set of tools and invokes music or effect generation based on a text description. Specific tool names, model weight sizes, supported output formats, and the license are not disclosed in the description — you'll have to dig into the repository itself for those. ### 💡 Why it matters Local inference for audio models is all about privacy and predictability: no limits on the number of requests, no billing for generation, and no prompts sent to someone else's server. Plus, the MCP wrapper embeds audio generation directly into the workflow of agentic clients like Claude Code, where previously you had to run a separate service with an API key. How smoothly it works in practice is a question for the repository itself: the author provides no benchmarks or VRAM/RAM requirements.