Scan Untrusted Text for AI Agents
Scan untrusted text for hidden controls, mixed scripts, encoded payload signals, and instruction-override patterns before an AI agent consumes it.
When to use it
Use this when an agent ingests text from webpages, email, support tickets, documents, comments, or retrieval systems. The scan is deterministic and looks at structural signals rather than trying to infer intent.
Important limitation
It does not prove content is malicious, safe, or a prompt injection. It gives the calling application specific indicators to review.
Use it in your browser
The browser version runs locally on your device.
Open the tool →Use it from an agent or application
POST https://api.scrubmytext.com/v1/scan
MCP tool name: scan_untrusted_text. Paid API tools use a bearer API key; inspect_text is also available on the free API tier.