StealthGPT API: humanization and agent workflows

Build AI humanization, writing and detection into your app with REST, async and bulk jobs, Super, Standard and Lite, or connect a compatible assistant through MCP.

Humanize an existing draft, generate new writing, score text for AI-generated patterns, or run a research-to-draft content workflow. Use REST endpoints in your app or publishing queue, or MCP tools for humanization and Stealth Agent in a compatible assistant.

​

Choose your workflow

Humanize existing text

Send source text with rephrase: true and choose Super, Standard or Lite. Use the model guide to select the amount of rewriting your workflow needs.

Generate from a prompt

Send rephrase: false for prompt-based writing. For multi-step research and drafting with academic, SEO or social presets, use Stealth Agent.

Run async and bulk jobs

Queue individual requests or batches of up to 100. Use polling or webhook callbacks to collect completed outputs, then route drafts for editorial review with the TypeScript and Python recipes.

Keep approved spans fixed

Replace links, citations, headings or other approved spans with markers in your application. Text preservation keeps those markers unchanged while the surrounding prose is humanized; restore the original spans afterward.

Connect an assistant through MCP

Discover humanization and Stealth Agent tools from a compatible client. Follow the client setup matrix and dated Codex verification for authentication, model selection and run retrieval.

Score text with the AI Detector

Submit text to the REST detection endpoint and use its documented response in your own review workflow. The detector is a separate endpoint from humanization.

​

Make your first request

Follow the Quick Start Guide and Authentication for REST setup. Then choose polling, callbacks or MCP, or start from the code examples. For n8n, Zapier and WordPress, see the integration guides in the sidebar.

API plans and web subscriptions are separate. Read Pricing for API entitlements, input/output word accounting and model-specific rates. Use Data Handling to plan request history, processing providers and account controls.

​

Inspect the output

The matched academic-writing study publishes original passages, first returned outputs and named detector results. The Super, Standard and Lite comparison shows concrete differences in rewriting, response times and final editing needs. Use these examples to choose a model, then review the result against your source material.

For integration errors, see Error Handling or contact [e-mail].