formant-studio
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@formant-studioExport my current session as a WAV file"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
FORMANT
FORMANT is a focused digital audio workstation for making short, expressive loops. It combines a polished step editor, playable Web Audio synthesis, a mixer, measured WAV export, and a local MCP server so an AI can read and edit the same session.
Six synthesized instruments, four editable bars, and one focused workspace keep the surface quick to learn while leaving room for detailed sound design.
Start the studio
Requires Node.js 22.12 or newer and a modern browser.
For the fastest setup, download the latest platform bundle from Releases. On macOS, choose FORMANT-macOS.tar.gz (the archive preserves the launcher's permission), open it, then double-click Install FORMANT.app. If macOS shows a security warning, Control-click the app, choose Open, then choose Open again. The Mac installer downloads a private Node.js 22 runtime automatically, so Node.js and Homebrew are not required. The Windows launcher can install Node.js through winget. The installer installs dependencies, builds the studio, and opens it in your browser.
npm ci
npm run build
npm run launchOpen http://127.0.0.1:4317. Windows: double-click Setup and Launch.cmd the first time; use Launch FORMANT.vbs afterward for a hidden background launch. macOS: double-click Install FORMANT.app in the downloaded bundle. If you prefer Terminal, run bash "Launch FORMANT.command". Nothing is installed as a startup service.
npm run stop stops only a verified FORMANT server at the configured address. npm start runs the server in the current terminal. The optional FORMANT_PORT environment variable selects another explicitly chosen loopback port; the app never evicts an existing listener or silently picks a different port.
Related MCP server: Pro Tools MCP Server
Make something
Press Play or Space. The seeded Midnight signal session is playable immediately. Browser audio requires that first click.
Select an instrument or a clip. Click the 16 pads to shape one bar. Adjust velocity below each step, or open Edit note & velocity for precise values and pitches.
Use the four Bar buttons to write variations. The copy button repeats the current pattern across all bars. Humanize varies active-step velocities; Clear clears only the current bar.
Shape each sound with Level, Pan, Tone, Decay. Use Mixer for all six channel faders. M mutes and S solos; mute takes precedence.
Set BPM, minor key, loop length and master volume. Swing is in Audio settings. Projects offers two starters, a blank session and project-file import.
Save project downloads a portable
.formant.json. Export audio renders the active bars plus a two-second release tail to 48 kHz stereo 16-bit PCM WAV. The render follows actual note, velocity, mute, solo, pan, tone and decay settings. Audio settings shows measured export peak, RMS, duration and file size.
Current edits are saved atomically to .formant/session.json. Undo/redo retains 100 edits during the service lifetime. Loading a project or preset is undoable. Closing the browser stops playback; reopening restores the saved session. No files or audio are sent to a cloud provider by FORMANT.
Connect an AI with MCP
Open AI connection → Copy configuration, then add that configuration to your MCP-compatible client. Or generate the exact local configuration:
node scripts/mcp-config.mjsThe MCP server uses stdio and the official TypeScript SDK. The local web service must be running. Keep a studio browser open for transport and export; session edits can run without a browser. Click Play once before asking an AI to start playback. A transport response acknowledges delivery; formant_get_session reports the browser's actual runtime.
Tool | Purpose |
| Read saved session, revision, history availability, browser runtime |
| Name, tempo, key, bar length, swing, master and loop |
| Instrument level, pan, tone, decay, mute, solo and name |
| Replace a zero-based bar with 16 velocities and optional MIDI notes |
| Set an event at index 0–63 |
| Request play, pause or stop |
| Midnight, Daybreak or empty session |
| Undo or redo an edit |
| Render a WAV through the browser and return a local file receipt |
| Read an export job's actual completion state |
Also includes formant://session and the produce_loop prompt. AI exports are confined to .formant/exports; no arbitrary paths, shell commands or credential access are exposed. Jobs are kept for the current server lifetime; completed WAV files persist.
Suggested first request: “Read my FORMANT session. Make the bass more syncopated, soften the hats, and set the tempo to 112. Read back the changes.”
Development and evidence
npm run typecheck
npm test
npm run test:mcp
node scripts/verify-mcp.mjs --exportThe MCP verifier makes temporary session edits and restores the captured session. Run it only in an idle test session. The export check needs a studio browser open. scripts/capture-qa.mjs captures independent Chromium viewports and runs Axe checks; install its matching browser with node node_modules/playwright/cli.js install chromium, or set FORMANT_BROWSER_PATH to an existing compatible headless Chromium executable. Runtime and test dependencies are locked in package-lock.json.
See docs/DESIGN.md, docs/QA.md, and the four independent vision reports in docs/reviews/. The screenshot reference is docs/concept.png.
This release is a synthesis-based groove workstation. Microphone recording, sample import, VST/AU hosting, arbitrary track creation and a built-in cloud language model are outside this release. The MCP interface lets your chosen AI collaborate. Windows runtime is verified; macOS source/launcher portability does not substitute for testing on a physical Mac.
Architecture and sources
React/Vite interface → one validated command reducer → atomic Node session store. Browser SSE updates provide shared state and explicit transport/export delivery. Look-ahead Web Audio scheduling and OfflineAudioContext rendering use the same instrument graph.
Implementation references: MDN Web Audio scheduling, OfflineAudioContext, and the official MCP TypeScript SDK server guide. See package-lock.json for the installed versions.
This server cannot be deployed
Maintenance
Related MCP Connectors
Create, co-edit, analyze, publish, and export collaborative step-sequencer sessions through MCP.
The team layer for AI coding agents: shared contracts, collision alerts, E2EE sessions.
- VocunoOAuthcom.vocuno
AI music studio: song generation with vocals, covers, stems, voice conversion, mastering, editing.
AI music production assistant — audio profiling, AI mixing sessions, and service inquiries.
Related MCP Servers
- AlicenseAqualityDmaintenanceEnables AI assistants to control Ableton Live through natural language by querying tracks, analyzing sessions, and exporting stems via AbletonOSC integration.91MIT
- AlicenseNot gradedqualityDmaintenanceEnables AI assistants to control Pro Tools via the PTSL gRPC API, providing session management, timeline navigation, track control, clip management, editing, markers, transport, audio analysis, and more.18MIT
- AlicenseNot gradedqualityCmaintenanceBridges AI assistants with Ableton Live, enabling real-time control, offline project analysis, version tracking, and rack/preset parsing for music production workflows.MIT
- AlicenseNot gradedqualityBmaintenanceEnables AI agents to control Ableton Live with full access to Live's object model, including clip creation, device control, automation, and audio signal capture for mixing and mastering tasks.1MIT