Put a deterministic text-integrity layer around your AI agent.
Language models are good at reasoning over text. They are not the ideal place to perform exact Unicode inspection, secret-pattern detection, or schema validation. ScrubMyText provides those mechanical checks through REST and MCP.
Before model ingestion
Use scan_untrusted_text, detect_secrets, or sanitize_agent_context on webpages, emails, documents, tickets, or retrieved context before it enters the model.
Before an action
Use validate_output to confirm that model-produced JSON satisfies the shape your application expects before a tool or downstream API receives it.
One-call pipeline
Use prepare_for_agent to normalize text, inspect characters, scan risk signals, detect credentials, redact patterns, and return agent-ready text plus a report.
Example workflow
email / webpage / document
↓
prepare_for_agent
↓
clean context + risk report
↓
AI agent
↓
validate_output
↓
downstream actionScrubMyText intentionally reports signals rather than claiming that text is “safe.” Semantic prompt injection and trust decisions remain the responsibility of the agent application.
Agent-facing tools
Every tool can be called over REST or the remote MCP server.
Scrub text
Remove hidden/control characters and normalize whitespace without rewriting wording.
Inspect text
See zero-width characters, bidi controls, Unicode normalization, and exact anomaly positions.
Scan untrusted text
Flag structural obfuscation, suspicious instruction patterns, mixed scripts, and encoded payload signals.
Redact sensitive text
Replace common credential, identifier, email, IP, SSN-shaped, and payment-number patterns.
Compare text
Verify exact changes and whether differences disappear after standard normalization.
Detect secrets
Find likely API keys, private keys, JWTs, cloud credentials, and other secret patterns without changing text.
Validate AI output
Parse JSON and validate common JSON Schema rules before an agent or application acts on output.
Sanitize agent context
Normalize untrusted context, redact sensitive patterns, and return risk indicators before model ingestion.
Prepare for agent
One-call preprocessing pipeline: scrub, inspect, scan, detect secrets, redact, and return agent-ready text.