Resonant MCP
MCP server for a Windows music studio: the agent writes lyrics, generates a song locally, arranges, mixes and exports WAV
- 174 ★ on the repository
- Last commit
- AGPL-3.0
Extended access
Why this level
- The agent reads and changes projects and WAV files in the chosen workspace and starts local GPU generation
- Voice conversion can imitate someone else's vocal, so the studio requires confirming rights to the reference
- Optional modules download a Python environment and model weights of several gigabytes
Install
- Works in
Manual install
{"mcpServers": {"resonant": {"type": "stdio", "command": "node", "args": ["${CLAUDE_PROJECT_DIR:-.}/mcp-dist/server.mjs", "--root", "${CLAUDE_PROJECT_DIR:-.}"]}}}The configuration from the repository for running from source. For the installed studio, Quick Start generates the configuration with the right paths.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The agent reads and changes projects and WAV files in the chosen workspace and starts local GPU generation
- Voice conversion can imitate someone else's vocal, so the studio requires confirming rights to the reference
- Optional modules download a Python environment and model weights of several gigabytes
This is third-party code. Review the repository files before installing.
What it does
Resonant is a free Windows studio with a clip launcher, linear arrangement, an instrument library and a mixer. It ships a local MCP server through which Codex, Claude Code or Cursor work on saved `.resonant` projects: they write lyrics in Songwriter Studio, program parts, place clips on the timeline, mix tracks, analyze the mix and export stereo WAV. Whole-song generation comes from the optional ACE-Step 1.5 module, which runs on your own computer. The agent can set direction, lyrics, tempo and duration, and can also pass a WAV as a timbre reference, a cover source or a passage to repaint. A second optional module, Seed-VC, transfers an isolated vocal take onto your own singing voice, and the agent must pass `voiceRightsConfirmed: true` for it. Writes to projects are confined to the folder you chose when connecting. Every change needs the current project revision, and the agent gets a bounded undo history. Audio is not returned to the language model. Songwriter Studio has six built-in language packs: English, Hindi, Mandarin, Korean, Spanish and Japanese.
Who it is for. Musicians and songwriters on Windows with an NVIDIA GPU who want to steer a project by voice through Codex or Claude Code and keep all the work on their own computer.
Good fit when
- You want a rough song from a single prompt: lyrics, arrangement, mix and a WAV file
- You have a vocal recording and want it moved onto your own singing voice with melody and timing kept
- A finished track needs a cover or a rewrite of one verse through ACE-Step based on a sample from the workspace
- The music is built from samples, SoundFonts and MIDI, and the agent places clips in scenes and on the timeline
- Cloud generators with subscriptions and quotas do not fit, and every file must stay on disk
Not a fit when
- You are on macOS or Linux: builds exist only for Windows 10 and 11
- Your GPU is weaker than an NVIDIA card with 6 GB of video memory: ACE-Step generation and voice conversion will not run on it
- You want another person's voice without their consent: the tool demands rights confirmation, and such use is not allowed
- You need a license for a closed commercial product: the code is under AGPL-3.0 and Seed-VC under GPL-3.0
Example request
Use Resonant to write an original alternative-soul song about rebuilding trust, generate a restrained vocal, leave mix headroom and export the final WAVMCP
- Transport
- stdio
- Authentication
- not required
Limitations
The prebuilt Windows builds are not signed yet, so SmartScreen shows a warning and the SHA-256 checksum from the release is worth comparing. Each of ACE-Step and Seed-VC downloads its own Python environment and model weights, which takes several gigabytes, with about 5 GB for Seed-VC after install. Voice conversion needs a separate vocal stem rather than a full mix. To run the MCP server from source, run `npm ci` and `npm run build:mcp` first.
How to disable. Remove the `resonant` block from the project's `.mcp.json` or `.codex/config.toml` and restart the agent. The studio itself is uninstalled through Windows installed apps; the ACE-Step and Seed-VC models sit in its folders and are deleted separately.
FAQ
How do I connect Resonant to Claude Code?
Open Quick Start in the studio, choose the assistant connection item, pick the workspace folder with your projects and copy the generated JSON into your project's `.mcp.json`. After a restart `/mcp` shows the `resonant` server as connected.
What can the agent do with my voice?
Seed-VC turns an isolated vocal into a take in your voice from a reference of 1 to 30 seconds. A run requires `voiceRightsConfirmed: true`, and the studio saves source and reference hashes next to the result.
Do I need internet or an account?
No. After the modules and models are downloaded, the studio works without internet, keys or a subscription. Downloads happen once, when you install ACE-Step, Seed-VC and instruments.
Which GPU do I need?
ACE-Step recommends NVIDIA with 6 to 8 GB of video memory; Seed-VC needs at least 6 GB and prefers 8. Before conversion the studio stops ACE-Step to free memory.
How does MCP mode differ from working in the studio by hand?
The agent sees the studio's features through `get_capabilities` and changes the project with focused calls. Each change is tied to a revision, so a simultaneous edit in the studio window will not be overwritten silently.
Discussion
No questions or reviews yet.
Related
MCP serversMCP server for REAPER: an agent creates projects, tracks and MIDI, adds effects, mixes, masters and renders a track
A local sound studio with an MCP server: an agent makes effects, music, songs and speech on open-weight models
A local MCP server generates music and sound effects, builds loops, adjusts loudness and exports audio for games or video.
MCP server for Audacity 3.x with 144 tools and 9 pipelines for audio cleanup, mastering and transcription, all processed locally.
BeatDesign MCP connects an agent to a local video editor, with clips, images and captions visible in the browser during editing.
Local speech synthesis for agents: a command and a skill voice text, clone a voice from a recording and render scores to audio, with no cloud
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| Resonant (this card) | no | not required | 174 | |
| Audio Studio MCP | yes | not required | 592 | a local MCP server for music, sound effects, loops and loudness without a full studio |
| Claude AI Music Skills | no | 630 | a Claude Code plugin for producing albums through Suno: concept, lyrics, prompts and mastering | |
| MCP Music Forge: track download and processing | no | key is optional | 5 | an MCP server that downloads and converts existing tracks rather than creating music |
README in short
The README presents Resonant as a local alternative to cloud song generators and walks from idea to finished WAV: lyrics, ACE-Step, vocal prep, Seed-VC, arrangement, mixing, export. It then covers MCP setup for Codex, Claude Code and Cursor, the instrument library, language packs and two install routes, and ends with system requirements and shortcuts.