Baseline every MCP tool. Detect risky drift. Enforce role-aware policy before execution. Scan responses for prompt-injection and secret leakage. Audit every allow, deny, monitor, and quarantine decision.

Built for teams deploying AI agents across multiple MCP servers in production.

Pre-release. Working with a small group of design partners— free for 90 days in exchange for honest feedback.

→ Talk to founder: [email protected] → Code: https://github.com/MaazAhmed47/Interlock

How Interlock Works

1. Interlock discovers MCP tools across your agent stack.
2. It normalizes tool metadata: effects, side effects, data classes, externality, identity mode, confidence, and warnings.
3. It stores a trusted baseline for each tool.
4. At runtime, Interlock inspects tool calls, arguments, role context, and responses before execution.
5. If a tool changes, Interlock detects drift, classifies severity, and can allow, monitor, deny, or quarantine.
6. Every decision is written to an audit log.

What Makes Interlock Different

Most agent security tools inspect one request at a time.

Interlock remembers your MCP tools over time.

It knows what each tool was supposed to do, detects when that behavior changes, and gives operators a safe review loop before agents can use risky new capabilities.

Who this is for