Visus Mcp

Built for production AI agents that need to fetch web content without eating prompt injection or leaking PII. Runs a Playwright-based pipeline that detects 19 categories of indirect prompt injection before sanitization, redacts emails and credit cards, and generates cryptographic proofs that the security pass actually ran. Handles PDF, JSON, and SVG in addition to HTML. The threat detection is granular: you get severity scores, confidence levels, and character offsets for each flagged attack. Designed around EU AI Act compliance requirements, with audit logs and Merkle-chained session ledgers. If your agent reads untrusted web pages and you need both safety and regulatory coverage, this is the tooling for it.