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

This advisory defines “Shadow MCP” — unapproved AI/agent server instances that can read/write internal systems — and explains how they proliferate, the risks they create (brand impersonation, internal data exposure, externally reachable endpoints), and a three-layer visibility model (external registries, external attack surface, internal inventory) plus five baseline questions organisations should answer to assess exposure.