Observed arrival · 2026-11-02
Perlecta: a hidden-text detector for agent inputs
An open-source MCP server that detects text hidden in PDFs, Word files, emails, and plain text before an AI agent reads the document.
- For
- Developers building document-reading AI agents
- Worth noticing
- The homepage lists 15 detectors and says file types are identified by file signature, not extension.
Field notes
The homepage demonstrates the detector on a sample garage repair quote: a line in invisible PDF render mode is absent from the printed view but extractable from the file. Perlecta's report gives an agent the anomaly code, page, coordinates, measurements, and a fingerprint, while a human-facing CLI output can reveal the excerpt as untrusted content. The page lists 15 detectors across PDF, Word, email, and plain text; it says file types are identified by signature, not extension.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
One card from the complete issue