Shadow MCP Servers: The AI Infrastructure You Can't See | UpGuard

This report defines "Shadow MCP"—unapproved, unsupervised AI/agent server instances that can read/write internal systems—and explains how they proliferate, the risks they pose (brand impersonation, data exposure, external endpoints), and three visibility layers (external registries, external attack surface, internal environment). It provides practical baseline questions and detection/governance guidance for mid-market security teams, recommending continuous registry monitoring, EASM integration for MCP discovery, and inclusion of MCP installation in procurement/policy workflows.