Code-driven game SFX with broadcast-style loudness mastering. Remotion, but for sound.
▶ Live demo · npm core · npm MCP
Generate jumps, coins, lasers, and UI ticks from code — or describe them in plain language and let your AI agent render them over MCP. Every sound is mastered to a per-category loudness target, so it sits in the mix without hand-tweaking gain.
In code — preset to WAV:
import { presets, renderSound } from "make-sfx-core";
import { writeFileSync } from "node:fs";
writeFileSync("jump.wav", renderSound(presets.jump)); // Uint8Array → WAVOr describe it — one MCP call generates WAV + reusable code:
sfx_from_description({ description: "coin pickup, a jump, and a higher-pitched blink",
outputDir: "./assets/sfx" })
npm install make-sfx-core # synthesis engine for your game
npx make-sfx-mcp # MCP server for Cursor / Claude (no install)- Describe → sound. Natural language becomes real WAVs through the MCP server — no DAW, no sample hunting.
- Loudness mastering built in. Each sound targets a category LUFS level (UI −14, combat −9, pickup −12, ambient −18, stinger −11) using ITU-R BS.1770 K-weighting and an oversampled true-peak limiter (≤ −1 dBTP). Bright coins and dark thuds end up equally loud to the ear, not just equal in RMS.
- Deterministic variation. A
seedyields reproducible micro-variants — same seed, same WAV, byte for byte — so builds and asset diffs stay stable.
| Feature | make-sfx | ZzFX | sfxr / bfxr / jsfxr |
|---|---|---|---|
| Parameter surface | 70+ (layers, LFOs, texture, FX) | ~22 (flat) | ~25–35 (GUI focused) |
| Synthesis model | Layered + modulators + texture + mastering | Single osc + fx | Single osc + limited fx |
| Loudness mastering | ITU-R BS.1770 K-weight + true-peak limiter (per-category LUFS) | None | None |
| Deterministic seeds | Yes (byte-identical variants) | Basic randomness | Random / mutate only |
| Code-first / library | Full TS configs, Node + browser | Yes (tiny) | Partial (some ports) |
| AI / natural language | MCP tools (sfx_from_description) |
— | — |
| Export | Mastered WAV + reusable code | WAV (browser) | WAV |
| Use case | Production game audio + AI agents | Tiny demos / js13k | Rapid retro prototyping |
Everything is a plain config object; the render pipeline is
layers → modulators → texture → master envelope → FX → loudness master → WAV.
Full parameter surface
| Group | Count | Highlights |
|---|---|---|
| Synth | 21 | volume, frequency, wave shape, ADSR, pitch slide, FM, tremolo, filter, bit-crush, delay |
| Master | 5 | duration (seconds!), tail, fadeIn, fadeOut, durationScale |
| OSC layers | 4 | mix, detune, delay, per-layer overrides |
| LFO modulators | 4 | amplitude / pitch / filter targets, rate, depth, shape |
| Texture | 11 | chorus, phaser, crackle, sub, transient, flutter, gate, grain, ringMod, wow |
| Post-FX | 12 | reverb, delay, 3-band EQ, compression, saturation, stereo width, limiter |
| Engineer knobs | 6 | pitchShift, brightness, warmth, punch, grit, space |
import { renderSound, presets } from "make-sfx-core";
renderSound({
params: presets.laser,
master: { duration: 0.3, fadeOut: 0.1 },
oscillators: [{ enabled: true, mix: 0.4, detune: 7 }],
modulators: [{ enabled: true, target: "pitch", rate: 12, depth: 0.4 }],
texture: { chorus: 0.3, phaser: 0.2 },
fx: { reverbMix: 0.3, eqHigh: 4, stereoWidth: 0.5 },
});Game presets (GAME_PACK_DESIGNS, same configs as MCP sfx_design_batch) ship extended
defaults — duration, layers, texture — so the describe workflow sounds good out of the box:
import { buildSoundConfig, GAME_PACK_DESIGNS, renderSound } from "make-sfx-core";
renderSound(buildSoundConfig(GAME_PACK_DESIGNS[0])); // fully-designed jumpjump blink coin hit explosion click powerup gameover laser
swoosh footstep alarm shield success error menu door reload engine
CLI:
npx make-sfx-core --preset jump --out jump.wav
npx make-sfx-core --preset blink --pitch-shift 1.5 --out blink.wav{ "mcpServers": { "make-sfx": { "command": "npx", "args": ["-y", "make-sfx-mcp"] } } }Workflow: sfx_from_description in one shot, or sfx_help → sfx_describe (preview) →
sfx_from_description (generate).
| Tool | Description |
|---|---|
sfx_from_description |
Primary — natural language → WAV + code |
sfx_describe |
Parse a description to configs (preview) |
sfx_game_pack |
Standard game pack + optional seed variants |
sfx_refine |
Refine an existing design with natural language |
sfx_help |
Workflow + examples |
sfx_get_parameters |
70+ params with ranges, units, aiHint |
sfx_list_design_profiles |
8 design starting points |
sfx_list_presets |
18 presets (tag filter) |
sfx_design / sfx_design_batch |
Manual full control (70+ params) |
sfx_create / sfx_batch_create |
Quick preset workflow |
sfx_export |
Export to a specific WAV path |
import { play } from "make-sfx-core/browser";
import { buildSoundConfig, GAME_PACK_DESIGNS } from "make-sfx-core";
play(buildSoundConfig(GAME_PACK_DESIGNS[0]));make-sfx-core— synthesis engine, 70+ params, loudness mastering,buildSoundConfig,GAME_PACK_DESIGNSmake-sfx-mcp— MCP server for AI agents (describe → WAV)@make-sfx/playground— web sound designer
pnpm install && pnpm build && pnpm test
pnpm dev:playground # sound designer at http://localhost:5173 (70+ live params)
pnpm example:generate # render assets/sfx/*.wav from GAME_PACK_DESIGNS
pnpm example:game # local demo page
pnpm demo:gif # render docs/demo.gif via Remotion (~15s)
pnpm publish:packages # maintainersLocal MCP dev (after pnpm build): point your config at
node ./packages/mcp/dist/server.js.
See CONTRIBUTING.md and the good first issues.
MIT — ZzFX synthesis algorithm © Frank Force (MIT), vendored in make-sfx-core.