{
  "slug": "mcp-human-in-the-loop-gate",
  "title": "Your MCP server can move money. Add the human gate in five lines.",
  "summary": "Tool poisoning and prompt injection are hard to stop. Unsigned side effects are not. How to require a human signature on dangerous MCP tool calls.",
  "lede": "Prompt injection is not going to be solved in general, and betting your refund tool on a classifier is a choice you will have to explain later. Authorization is solvable. Here is how to make the dangerous tools on your Model Context Protocol server unable to execute without a signature that no amount of injected text can produce.",
  "date": "2026-09-14",
  "reading_time": "18 min read",
  "category": "Agents",
  "tags": ["MCP", "Model Context Protocol", "tool poisoning", "prompt injection", "human in the loop", "agent security", "AI agents"],
  "image": "https://cdn.twc.sh/images/igcache/MCP%20Human%20Gate/1200_630/blog.jpg",
  "url": "/blog/mcp-human-in-the-loop-gate.html",
  "wordcount": 3756,
  "related": ["approval-theater", "mcp-identity-12-lines", "agent-deleted-production"],
  "schema": "Article"
}