Docs

Sign in to mint an API key, then point the MCP server at it. Every generation returns a durable URL you can drop straight into your app.

{
  "mcpServers": {
    "media": {
      "command": "node",
      "args": ["/abs/path/to/apps/mcp/dist/index.js"],
      "env": {
        "MEDIA_API_URL": "https://your-api.workers.dev",
        "MEDIA_API_KEY": "mk_your_key_here"
      }
    }
  }
}

curl -X POST $MEDIA_API_URL/v1/generate \
  -H "authorization: Bearer $MEDIA_API_KEY" \
  -H "content-type: application/json" \
  -d '{"type":"speech","brief":"Your order is on its way.","voice":"narrator"}'

Six audio types, powered by ElevenLabs. Speech is spoken exactly as written — the engine never rewrites a script. Sound effects and music get the same intent presets and prompt enhancement as images.

Sound packs

One call renders every clip an app's interaction layer needs, in a single style, and returns a manifest.json mapping name → URL. Your brief is the shared style (“soft glassy minimal”, “warm retro arcade”).

curl -X POST $MEDIA_API_URL/v1/generate \
  -H "authorization: Bearer $MEDIA_API_KEY" \
  -H "content-type: application/json" \
  -d '{
    "type": "soundpack",
    "brief": "soft glassy minimal, premium feel",
    "pack": "ui-essentials",
    "audioFormat": "wav"
  }'

# → 8 clips + manifest.json, every clip peak-normalized to -1 dBFS

Choosing a voice

voice accepts a built-in alias, a catalog name, or a raw voice id. Aliases include narrator, assistant, guide, calm, friendly, announcer, hero and villain. Browse everything with GET /v1/voices. To give your app its own voice, design one and reuse the id it returns:

# 1 — design (returns 3 candidate previews, saves the first)
{"type":"voice",
 "brief":"warm, upbeat female product guide, late twenties, neutral American accent",
 "saveVoiceAs":"Acme Guide"}
# → result.meta.voiceId

# 2 — every line from then on
{"type":"speech","brief":"Welcome back!","voice":"<that voiceId>"}

Formats

mp3 (default, 44.1 kHz), wav (24 kHz PCM, peak-normalized — best for game engines and Web Audio), pcm, opus, ulaw. Level consistency across a pack requires a linear format, so reach for wav when clips must sit together.

All /v1 routes take authorization: Bearer <api key>. Jobs run asynchronously; a completed job carries result.assets[] (durable URLs) and result.meta (voice used, pack name, manifest URL).