ask-human-mcp

open source

Zero-config MCP server for human-in-loop validation to prevent AI agent hallucinations

Updated 2025-01-15

Tool EcosystemsFree· Open source, self-hosted via pip installVisit site Docs

ask-human-mcp is an MCP (Model Context Protocol) server that pauses AI agents when they encounter uncertainty, logs questions to a markdown file, and resumes execution once a human provides the answer. It solves a critical pain point for solo AI builders: preventing agents from confidently generating hallucinated endpoints, incorrect assumptions, or misinterpreted code logic that leads to hours of debugging.

Use Cases

·Prevent AI agents from hallucinating API endpoints or authentication flows
·Validate architectural decisions before code generation
·Clarify ambiguous requirements during agent execution without manual rewrites
·Build trustworthy agent workflows for production-critical systems

Key Points

·Zero-config integration: single pip install + one-line .cursor/mcp.json addition
·Markdown-based Q&A: questions logged to ask_human.md with PENDING status, human fills in answer, agent resumes
·Cross-platform support: macOS, Linux, Windows with no external servers or webhooks required
·File safety: built-in file locking, automatic rotation at 50MB, prevents corruption from concurrent access
·Lightweight escape hatch: agent calls ask_human(question, context) when uncertain, avoiding costly hallucinations

ask-human-mcp

Use Cases

Key Points

Tags