stt_create dry_run and max_spend_usd: cap an MCP transcription run

On the Sume hosted MCP, call stt_create with dry_run: true first to preview admission and cost without submitting a job, then call it again without dry_run to submit. Add max_spend_usd when you want Sume to enforce a ceiling; it is applied only when you pass it. Every paid write needs its own idempotency_key. These gates are documented in MCP tools and gates, read 2026-10-06.

What each argument does

The tool also accepts the STT request fields (audio_url, language_code, duration_seconds, segmentation, metadata). Send duration_seconds so the preview prices seconds rather than the one-minute default.

The two calls

The first call is the preview. If the cost and queue room are acceptable, send the same arguments without dry_run.

{
  "name": "stt_create",
  "arguments": {
    "audio_url": "https://media.sume.com/example/clip.wav",
    "duration_seconds": 12,
    "idempotency_key": "stt-clip-0001",
    "max_spend_usd": 0.01,
    "dry_run": true
  }
}

Previewing a whole batch

For many clips, the generation_admission_preview tool answers queue and balance questions for a request shape in one call, and usage_get reads the ledger afterward. Pair them with a jobs_wait on the wave.