Skip to main content
Stealth Agent and humanization runs are asynchronous: you start a run, get a runId immediately, and the long-running work (research, drafting, humanization, scoring) happens in the background. There are three recommended patterns for consuming it.

​
1. Queue and poll

Best for scripts, cron jobs, and simple backends.
  1. Start a run: POST /api/stealthify/runs (humanize-only) or POST /api/stealthify/agent/runs (full pipeline) → 202 with runId + statusUrl.
  2. Poll GET .../runs/{runId} every 5–15 seconds until status is completed, failed, or cancelled.
  3. Read the final result and billing metadata from the completed response.
Use an idempotency-key header on create so retries never enqueue duplicate runs or double-charge. → Humanizer async runs · Stealth Agent async runs

​
2. Fire and webhook

Best for no-code tools (n8n, Make, WordPress) and event-driven backends that don’t want to poll.
  1. Start a run and include webhookUrl (and optionally webhookSecret).
  2. When the run reaches a terminal state, StealthGPT POSTs the result (or terminal failure) to your webhookUrl.
  3. Verify the x-stealthgpt-signature HMAC (when webhookSecret is set) and reconcile by x-stealthgpt-run-id.
Webhooks are best-effort with retries on 429/5xx; treat delivery as at-least-once and dedupe by run id + status. → See the Webhook callbacks section on the async runs page.

​
3. Agent / tool invocation

Best for AI agents and chat clients — let the model call StealthGPT as a tool via our MCP server (Cursor today; claude.ai web & ChatGPT via OAuth coming soon). → Use Stealth Agent from MCP clients

​
Picking a pattern