Mcp As A Judge

This is a validation layer that sits between your AI coding assistant and the LLM, acting as a gatekeeper for code changes. It exposes tools like judge_coding_plan, judge_code_change, and judge_testing_implementation that use MCP sampling to evaluate whether your research is thorough, your diffs meet engineering standards, and your tests actually work. It also includes raise_missing_requirements and raise_obstacle for interactive decision making via MCP elicitation. You'd reach for this if you're tired of AI assistants hallucinating APIs, skipping proper testing, or making unilateral architectural decisions. Works best with GitHub Copilot in VS Code where sampling is native, though other assistants need an LLM API key configured. Think of it as enforcing a code review process before the AI's changes land.