Free sounds via MCP & API for your video game

Give a q, at least one filter, or both. Durations are in seconds.

With type=music, full mixes come first, then loops, then single parts, unless you set parts. good_for moves sounds suited to a job (transition, ui_click, background_bed…) up the results without filtering any out. A search with q that finds nothing within its type searches every type and lists type in relaxed_filters.

A page of results, with a link to the next page. bpm and key describe music; creator is null when a sound has no public creator.

{
  "page": 1,
  "per_page": 5,
  "total_results": 38,
  "next_page": "https://api.soundza.com/v1/sounds/search?q=whoosh&type=SOUND_EFFECT&per_page=5&page=2",
  "search_id": "x7k2m9q4r1t8v3w6y0z5a2b4",
  "sounds": [
    {
      "id": "cm9w2k4e30001",
      "url": "https://soundza.com/sound/fast-air-whoosh",
      "title": "Fast Air Whoosh",
      "description": "A quick, bright air whoosh that sweeps left to right.",
      "creator_description": null,
      "type": "sound_effect",
      "form": null,
      "parts": null,
      "vocals": null,
      "style": null,
      "good_for": ["transition"],
      "duration": 0.84,
      "bpm": null,
      "key": null,
      "chord": null,
      "note": null,
      "mood": ["airy", "bright"],
      "tags": ["whoosh", "swipe", "air", "airy", "bright"],
      "sync_ms": 310,
      "lufs": -19.4,
      "true_peak_db": -6.1,
      "loudness": { "value": -18.2, "basis": "momentary_max" },
      "gain_db": 5.1,
      "loop_seamless": null,
      "file_type": "audio/wav",
      "file_extension": "wav",
      "sample_rate": 48000,
      "bit_depth": 24,
      "channels": 2,
      "size_bytes": 322604,
      "category": { "id": "WHSH", "label": "Swooshes › Whoosh" },
      "pack": {
        "id": "cm8p1x7a20004",
        "name": "Air and Motion",
        "slug": "air-and-motion",
        "sound_count": 42,
        "url": "https://soundza.com/packs/air-and-motion"
      },
      "creator": null,
      "preview_files": [
        { "type": "audio/mpeg", "bitrate": 128, "link": "https://media.soundza.com/l/fast-air-whoosh/4fJ9aQx2LmP0sT7vKc1Rb8Ze.mp3" }
      ],
      "download_url": "https://api.soundza.com/v1/sounds/cm9w2k4e30001/file",
      "license": { "name": "Soundza Standard License", "url": "https://soundza.com/license" },
      "attribution": "\"Fast Air Whoosh\" on Soundza (https://soundza.com/sound/fast-air-whoosh)"
    }
  ]
}

Files are level-matched, so sounds sit together without adjusting each one: one gain per sound, no compression, peaks at or under −1 dBTP. Ambience aims for −20 LUFS, music, pads and long loops for −16, everything else for −13 on its loudest 400 ms. lufs, true_peak_db and loudness describe the original; gain_db is what the default file adds to it.

Add ?variant=original for the original as uploaded. The redirect is a signed link that lasts 5 minutes, so keep the file, not the link. download_url already carries the search it came from; if you serve many users, add user with your own id for each.

A free key may keep up to 500 downloaded sounds. Now and then a sound is taken down; check GET /v1/removals at least monthly and delete any it lists.

prompt up to 200 characters. duration_seconds0.5 to 10, default 3: it's for sound effects, so for music use generate_music below. takes 1 to 3, best first, default 1. seed to repeat a result. Each take has a WAV url and an MP3 preview_url; neither needs a key, and both work for 120 days, so save what you keep.

250 a week per account (at most 120 an hour), one at a time. A busy moment answers 503 or 429 with Retry-After: wait and send it again. A sound that takes longer than about 50 seconds answers 202 with a poll_url; send "wait": false to always get that.

Takes are made with AI and say so in ai_content. Use them like any soundza sound. They aren't exclusive to you, and we may add good ones to the soundza library for everyone (see AI features in the Terms).

A sound_id is all it needs, but spread, how different the variations are, is what makes the set fit how the sound is used. Use subtle for sounds that must stay consistent, like UI clicks, coins and gunshots; natural for footsteps, impacts and swings; and bold when each take should be a different performance, like grunts from different characters. All variation parameters lists every field and its type; here's a full example:

# Short sounds come back in 5-10 seconds, in this one call.
curl -s -X POST "https://api.soundza.com/v1/variations" \
  -H "Authorization: Bearer $SOUNDZA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "sound_id": "footstep-on-wood",
    "spread": "natural",
    "takes": 6
  }'

# then save each take under its own name:
#   curl -sL -o footstep-on-wood-var-01.wav "<takes[0].url>"

Working through an agent? Tell it what the sound is for, and it picks the level and the number of takes for you.

spread goes from 1, about as different as a pitch randomizer, to 10, the same idea played by someone else. It's measured on each sound, so a spread of 5 sounds about as different on a punch as on a footstep. The presets are "subtle" (2), "natural" (5), "bold" (8).

Every variation comes back as a WAV (url) and an MP3 preview (preview_url), named after the sound: footstep-on-wood-var-01.wavand on. Each one is lined up with the original's attack and matched to its length, tone and loudness, so you can lay them along a video timeline, or drop the set into Godot's AudioStreamRandomizer, an FMOD or Wwise random container, or Unity next to the original. Links work for 120 days, so save what you keep.

It works on one-shots up to 10 seconds: sound effects, instrument hits and stingers. Each variation counts as one of your 250 generations a week, free, so 6 takes use 6. Short sounds come back in 5 to 10 seconds; a 10-second sound can take up to about 30.

If the service is busy, you'll get a 429 or 503 response telling you how long to wait before trying again.

A plain description works, but game, scene and moodare what make the music feel like it belongs in your game — the more you give, the better the fit. All music generation parameters lists every field and its type; here's a full example:

# Music takes a minute or two, so this answers 202 with a poll_url.
curl -s -X POST "https://api.soundza.com/v1/music" \
  -H "Authorization: Bearer $SOUNDZA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "description": "a slow, warm orchestral piece: solo harp, soft strings, no drums",
    "looping": true,
    "game": "a cozy fantasy RPG about running a village inn",
    "scene": "the player wanders the village at dawn",
    "mood": "peaceful and welcoming"
  }'

# then poll until it is ready:
curl -s "https://api.soundza.com/v1/music/JOB_ID" -H "Authorization: Bearer $SOUNDZA_API_KEY"

Working through an agent? Tell it about your game and the scene, and it fills the fields in for you.

The endpoint supports two models: stable (Stable Audio 3) and lyria (Lyria). We recommend using stable for loops and short tracks, or lyria for long, full-length tracks. Only stable supports looping.

If you need a seamless loop, use stable with looping: true. We do some extra audio work on the back end to make that possible, so you get a clean loop that plays back to back with no gap. You can run 15 stable and 8 lyria generations a week, free, one at a time.

Every generation comes back as a WAV (url) and an MP3 preview (preview_url). Use the WAV in your game: it loops with no gap and includes loop points for Unity and Godot. The MP3 is there for a quick listen and has a short gap if you loop it. Loops also report their tempo and bar count.

If the service is busy, you'll get a 429 or 503 response telling you how long to wait before trying again.

Every response has X-RateLimit-Limit, X-RateLimit-Remaining, X-RateLimit-Reset and X-RateLimit-Policy. A 429 also has Retry-After, in seconds.

Only different sounds count as downloads. Total file fetches are capped at twice the download limits, so save each file once and reuse it by the sound's id. Generations are at most 120 an hour, one at a time, each up to 10 seconds. Looking up a single sound or pack has its own limit, five times the search limit.

Trying it without a key: 20 searches a minute and 300 a day per connection, 3 pages of 20, and no downloads.

We are flexible on how you credit us, but we ask that you please do so somehow that works naturally for your project.

Want to use our logo? You're welcome to show it with your link: the full logo (PNG) or the mark on its own (SVG). Just don't use it as your app's icon.

• Once people outside your team see what you make, link back to soundza.com in the way that fits it: "Powered by soundza" by your sound picker plus a link on your site, a thank-you wherever you share your game, or a line in your video description. See soundza.com/developers#credit.
• Request a sound's download_url when it is added to a project, save the file once, and reuse it by the sound's id.
• Previews are for listening. Use downloaded files in finished work.
• Don't copy Soundza's core functionality or build a competing sound library or API.
• A free key may store up to 500 downloaded sounds. Remove any sound listed in GET /v1/removals, and check it at least monthly.
• Respect the rate limits, and ask when you need more. We may suspend or end API access for anyone, at any time, for any reason.

Errors are JSON: {"error": {"code", "message", "details"?}}.

Two server routes that keep your key private, and a component with chips, previews and the credit link. Style it to match your app.

// app/api/sounds/route.ts: search, with your key kept on the server
export async function GET(req: Request) {
  const params = new URL(req.url).searchParams;
  const res = await fetch(`https://api.soundza.com/v1/sounds/search?${params}`, {
    headers: { Authorization: `Bearer ${process.env.SOUNDZA_API_KEY}` },
  });
  return new Response(res.body, { status: res.status, headers: { "Content-Type": "application/json" } });
}

// app/api/sounds/[id]/file/route.ts: call it from your server when a user adds a sound
export async function GET(req: Request, { params }: { params: Promise<{ id: string }> }) {
  const { id } = await params;
  const query = new URL(req.url).searchParams; // search_id, and your own user id as user
  const res = await fetch(`https://api.soundza.com/v1/sounds/${encodeURIComponent(id)}/file?${query}`, {
    headers: { Authorization: `Bearer ${process.env.SOUNDZA_API_KEY}` },
    redirect: "manual",
  });
  const location = res.headers.get("location");
  return location ? Response.redirect(location, 302) : new Response(res.body, { status: res.status });
}

When a user adds a sound, fetch the file route from your server and save the file to your own storage. That counts the download. The route redirects to a signed link that lasts 5 minutes and can't be read from a browser on another site, so don't fetch it client-side or keep the link. Pass your own user id as user if you want per-user reports.

The MCP server has these tools: search_sounds, get_sound, search_packs, get_pack, download_sound, generate_sound, vary_sound, get_generation, generate_music and get_music.

REST

https://api.soundza.com/v1

MCP, sign-in

https://mcp.soundza.com/mcp/oauth

MCP, API key

https://mcp.soundza.com/mcp

OpenAPI

soundza.com/openapi.json

For agents

soundza.com/llms.txt