Context First Mcp

A 37-tool suite built around `context_loop`, an orchestrator that runs eight checks per turn: recap, conflict detection, ambiguity scanning, entropy monitoring, abstention logic, tool discovery, drift detection, and synthesis. Returns a single directive with action type, context health score, extracted facts, and next-tool suggestions. Includes persistent memory with graph queries, five advanced reasoning engines (InftyThink, Coconut, ExtraCoT, MindEvolution, KAG-Thinker), and seven truthfulness verifiers including neighborhood consistency checks and self-correction. Also ships sandbox quarantine for isolating sub-task contexts and a structured research pipeline. Reach for this when long conversations start contradicting themselves or when you need the AI to ask clarifying questions instead of guessing. Works over stdio or as a Vercel-deployed HTTP endpoint.