Skip to main content
AI Interview Question
All Questions
DEEP EXPLANATION

Human Approval Hooks for Dangerous MCP Tools (EXPLAINED)

Scenario BasedMCPHard22 min read

TL;DR — Quick Answer

Enforce approval at the gateway (authoritative policy) with host UI for prompts; dangerous tools enter pending state until approver confirms out-of-band identity; audit every decision; model receives PENDING_APPROVAL not silent success.

The Interview Question

Design a human-in-the-loop approval flow for destructive MCP tools (prod deploys, IAM changes, bulk deletes). Where does approval live — host, gateway, or server?

Deep Explanation

Sign in to unlock full answer

Get deep explanations, PDF export & all MCP questions

  • 18 more sections of deep explanation
  • Real-world examples
  • Common mistakes
  • Interviewer expectations
  • Follow-up questions
MCPHuman-in-the-LoopApprovalSafetyGoogleAmazonMicrosoft