Audio Studio MCP

What it does

Community MCP server that runs Stability AI's open Stable Audio 3 weights on your own machine — MLX on Apple Silicon, PyTorch on NVIDIA — to generate instrumental music beds, stems and sound effects from text, 1–4 seeded candidates per call. It goes past generation: edit_audio re-styles a clip, regenerates a time range or extends a track, and process_audio trims, loops, fades and normalizes to a LUFS target before writing WAV, OGG or FLAC with a JSON sidecar recording prompt and seed. For a filmmaker-agent that is a free, reproducible temp-score and foley-draft loop that never uploads the cut.

Connect

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

Local install: Multi-step: git clone https://github.com/kanaworksai-hue/audio-studio-mcp, then ./install.sh (macOS Apple Silicon) or install.ps1 (Windows + NVIDIA). The installer sets up uv and a .venv, vendors the Stable Audio 3 runtime, downloads the medium model and prints the exact claude mcp add line with your path.

Sample tools

Field notes

Five days old when listed: the repo was created 18 Oct 2026 and reached v0.2.2 on 22 Oct with 447 stars and 33 forks, so the star count is launch-week attention, not a track record. MIT, one maintainer, no PyPI package — install is a clone plus installer script. Hardware is the real gate: Apple Silicon (16 GB RAM recommended, ~6 GB for the medium model) or Windows with an 8 GB+ NVIDIA GPU (~14 GB with PyTorch); Linux with NVIDIA is untested and has no installer, Intel Macs and AMD GPUs are unsupported, and ffmpeg must be on PATH. Instrumental only — Stable Audio 3 cannot do vocals or lyrics — and the README says timing cues like 'climax at 15 s' are not reliably followed, so cut to picture with inpaint and continue rather than prompting for hits. Outputs fall under the Stability AI Community License, which allows free commercial use only below US$1M annual revenue; above that you need a Stability licence before anything ships. Not affiliated with Stability AI. For sync-to-picture scoring from licensed catalogues use `sonilo`; for hosted SFX, ElevenLabs or the fal MCP. Compared with `audiogen`, this is the better-maintained and broader offline option.

Also in Sound Design & SFX

Mirelo MCPMirelo AI Official

Foley generated against the picture, then extended or patched region by region.

sfx · video-to-sfxremote · credits

AudioGen MCPpeerjakobsen

Local, offline sound-effect generation with Meta's AudioGen — no API key, no credits.

sfx · local-inferencefree

Auphonic (via Zapier MCP)Auphonic / Zapier

Auphonic's automated audio post (loudness, leveling, denoise) reachable through Zapier MCP.

audio-cleanup · loudness-normalizationremote · freemium

A AudioLab MCPAudioLaunch Official

Loudness, true-peak and voice metering that grades a mix against real delivery specs.

loudness-analysis · true-peakremote · freemium

REAPER MCPbonfire-systems / Youssef Hemimy

Drives a live REAPER session: tracks, MIDI, FX, mixing, stems, and loudness analysis.

daw-control · mixingfree

Freesound MCP Serverjohnkimdw

Search freesound.org's CC-licensed sound library by natural language for SFX pulls.

sfx-search · audio-metadatafree

Pairs well with

A Artyfile Music LicensingArtyfile LTD Official

Search a catalogue of real recordings and buy a lifetime worldwide sync licence per track, WAV plus certificate.

music-search · music-licensingremote · paid

AudioPod MCPAudioPod AI Official

Hosted MCP for TTS, voice cloning, music, stem and speaker separation, and denoise.

tts · voice-cloningremote · freemium

Cartesia MCPCartesia Official

Official Cartesia (Sonic) server: low-latency TTS, cloning, voice change, localization.

tts · sttremote · freemium