Show HN: An MCP server that finds text your agent reads and you can't see

Category: security

Tags: mcp, security, document-analysis, ai-agents, prompt-injection

Score: 6.0/10 (Innovation: 6, Technical: 5, Documentation: 6, Utility: 7)

HiddenContent is an MCP server that detects invisible or hidden text in documents (PDF, DOCX, HTML, etc.) before an AI agent reads them, mitigating prompt injection and data exfiltration risks. The detection engine is closed-source and API-hosted, with this repository only containing distribution bundles. It addresses a real security gap in agent workflows, though the closed nature and token requirement limit openness.

Target audience: AI engineers, security-conscious developers, and teams deploying LLM agents

Repository: https://hiddencontent.ai/start/ · MIT

View on Hacker News