Ask Human

This MCP server enables AI agents to escalate questions to humans through a markdown file-based workflow, preventing hallucinations by providing a direct human-in-the-loop mechanism. Built by Mason Yarbrough for Kallro Inc. development workflows, it writes AI questions to a markdown file that humans edit with answers, using file watching to detect responses and return them to the AI agent. The implementation features cross-platform file locking, atomic operations, question timeout handling, file rotation for large logs, and robust markdown parsing, making it valuable for development scenarios where AI assistants need authoritative human input on decisions like database passwords, API endpoints, or architectural choices rather than generating potentially incorrect responses.

Source

Repository: https://github.com/masony817/ask-human-mcp