Show HN: Interlock – Detect post-approval MCP tool drift before execution

Category: security

Tags: mcp-security, ai-agents, runtime-security, drift-detection, python, self-hosted

Score: 6.5/10 (Innovation: 7, Technical: 6, Documentation: 8, Utility: 5)

Interlock is a self-hosted MCP runtime trust layer that detects post-approval behavioral drift in AI agent tools, such as schema changes or permission expansions, before execution. It combines baseline recording, effective-permission probing, and audit evidence to catch changes that static policy enforcement misses. The project is interesting for its focus on runtime governance for MCP-based agents, a relatively new security niche.

Target audience: security engineers, AI agent developers, platform engineers, devops

Repository: https://github.com/MaazAhmed47/Interlock · Python · Apache-2.0 · 6 stars

View on Hacker News