Deep Analyst Prover MCP Connector for Claude

The goal is to produce analysis that a domain expert would find genuinely insightful, not a structured restatement of the question. You must: (1) DECOMPOSE — break the problem into 3-5 atomic sub-problems using First Principles. Strip convention: "we do it this way because everyone does" is not an answer. Each sub-problem must be independently analyzable, (2) SURFACE ASSUMPTIONS — list 3-5 load-bearing beliefs with consequences if wrong. "If this assumption is wrong, EVERYTHING collapses." Focus on: stability assumptions (what will not change?), motivation assumptions (how will people behave?), prediction assumptions (what will happen?), (3) MULTI-MODEL — apply 3+ named mental models: First Principles, Second-Order, Inversion, Opportunity Cost, Circle of Competence. Where models DISAGREE is where the real insight lives. Single-lens analysis is systematic blindness (Munger), (4) STEELMAN — make the STRONGEST case for the opposing view. Ideological Turing Test: could the opponent read your steelman and say "yes, that IS my argument"? Not "some might argue..." — that is a strawman, (5) CASCADE — "and then what?" 3 levels. L1 immediate → L2 what does L1 cause → L3 systemic. 10/10/10 (Welch): what happens in 10 minutes, 10 months, 10 years? Most stop at L1. See L3, (6) PREMORTEM — "It is 12 months later. It failed spectacularly. Why?" (Klein 2007). 3+ specific, plausible, non-obvious failure paths. Prospective hindsight makes risks 30% more identifiable, (7) SYNTHESIZE — combine ALL models into a NOVEL insight that could not emerge from any single model. Synthesis ≠ summary. If your conclusion could apply to any problem, it is not a conclusion. If rejected, your analysis is surface-level. Structured reflection tool for deep, multi-model intellectual analysis — writing, strategy, decisions, research, brainstorming. Stacks 5+ mental models (Munger, Kahneman, Klein) against the same problem: First Principles decomposition, explicit assumptions, Second-Order cascades (3 levels deep), Steelmanning (Ideological Turing Test standard), and Premortem risk mapping. Catches Surface Analysis (restating the question as the answer — "the key challenge is the challenge" without decomposition into atomic sub-problems that can be independently analyzed), Hidden Assumptions (load-bearing beliefs never made explicit — "users want speed" when the real bottleneck is trust, but the speed assumption was never questioned because it felt obvious), Single-Model Thinking (analyzing through one lens only — "economically optimal" without considering behavioral, political, ethical, or temporal dimensions. Single-lens = systematic blindness), Strawman Opposition ("some might argue..." — presenting a weak version of the counterargument. Steelman: the STRONGEST case against your position, to Ideological Turing Test standard), Cascade Blindness (seeing only L1 consequences — "this will increase efficiency" without tracing L2 "which team gets cut?" and L3 "what institutional knowledge is lost forever?"), and Generic Conclusions (a synthesis that could apply to any problem is not a synthesis — "we should be thoughtful and move carefully" says nothing). Call once per complex analysis, strategic decision, or research synthesis