This product is an MCP server that scans documents for concealed or invisible text and other forms of hidden content before an AI assistant reads or acts on them, returning a verdict (e.g., "clean" or "suspicious"), findings, and a signed certificate. It catches things that plain-text checks miss - white-on-white text, tiny fonts, tracked changes, text behind images, and other file-level concealment techniques - and reports specific techniques (like docx.vanish) and excerpts when present. The service exposes an API (/v1/analyze) and connectors for agent platforms so the check can run automatically when a file is shared, or be invoked from code in a pipeline; responses include a scope indicator that signals whether a full file check was possible or only a text-level partial check.
Sign-up is via GitHub (20 free pages) or an API call; keys are shown once and billed per page (roughly $0.01 per page, $10 minimum for 1,000 pages), with Stripe checkout and optional auto-refill. Integrations include Claude (web/mobile and desktop), ChatGPT (requires Developer mode/Plugins to get full file analysis), Claude Desktop local extension/skill, and MCP-capable clients like Cursor; a desktop-only local server mode is available to keep files on your machine. The docs show curl examples, connector setup, key management, usage calls, and emphasize that files are analyzed on their servers (no third-party model calls claimed).
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.