Glossary · MCP

Tool rug pull

A tool rug pull is an attack where a tool’s definition changes after it was reviewed and trusted. The version an operator approved was benign; a later update quietly swaps the description or parameters for malicious ones. Point-in-time review cannot catch it, because the pull happens after the review.

Rug pulls weaponize the trust earned by good behavior, the same dynamic as compromised software dependencies. The defense is continuous verification: re-screen tool definitions at every connect, and pin what was approved so any change surfaces as a governed event requiring re-approval rather than taking effect silently. Trust the first sight after screening; verify every sight after that.

Related

Shrike governs AI agent actions in real time: every command, query, and tool call evaluated against policy before it executes. Start with what is action governance or the quickstart.