Connects Claude to structured adversarial analysis via four modes: counter (default tenth man protocol that attacks your decision), probe (pointed questions across motivation/evidence/alternatives/stakes/timing), redteam (ranked failure scenarios with likelihood and cascades), and premortem (writes the postmortem from 12 months in the future). Takes a decision statement plus optional context, returns engineered prompts that your LLM processes locally. No API keys, no external calls. Useful when you need to stress test architecture choices, strategy pivots, launch plans, or hiring decisions before consequences surface. Works through a single analyze tool with mode parameter. Quality scales with your model's capability since it's doing prompt construction, not execution.
claude mcp add --transport stdio bluwork-kontra-mcp uvx kontra-mcp