Mental Models MCP

Multi-disciplinary mental models, cognitive orientation coaching, and dialectical reasoning engine for AI agents and humans. Zero API keys required.

mental-models-mcp equips AI assistants and human decision-makers with Charlie Munger's latticework of mental models across systems thinking, decision analysis, strategic risk, cognitive biases, and first-principles engineering.

Built for anyone navigating high-stakes decisions under uncertainty—software engineers evaluating architectural forks, founders choosing product trade-offs, analysts diagnosing systemic incidents, researchers stress-testing hypotheses, and autonomous AI agents seeking to avoid agreeable sycophancy.

Quickstart

Option A — Any MCP Harness (incl. Claude Desktop, Cursor, Antigravity, Windsurf)

Add to your client configuration:

{
  "mcpServers": {
    "mental-models": {
      "command": "uvx",
      "args": ["mental-models-mcp"]
    }
  }
}

Or via NPX:

{
  "mcpServers": {
    "mental-models": {
      "command": "npx",
      "args": ["-y", "mental-models-mcp"]
    }
  }
}

Option B — Claude Code CLI Plugin

claude plugin marketplace add surendranb/mental-models-mcp
claude plugin install mental-models-mcp

Option C — Standalone CLI Execution

uvx mental-models-mcp

Core Tools

1. route_mental_model(problem, domain, problem_type): Lightweight L0 index scoring candidate frameworks against your situation. Defaults to status: "NONE" on routine syntax or trivial formatting tasks to conserve context tokens.
2. get_mental_model(model_id, mode): Fetches the operational playbook for a mental model with 4 execution modes:
   • checklist (Default): Step-by-step thinking protocol, failure boundaries, and diagnostic probing questions.
   • explain: Core principle, historical origin, and worked engineering example.
   • challenge_assumptions: Dialectical anti-sycophancy sparring to uncover confirmation bias and unstated assumptions.
   • full: Complete unabridged markdown playbook.
3. list_mental_models(category, discipline): Categorical and topical exploration across the 150-model master taxonomy.
4. skills_list() / skill_read(name): Dynamic runtime onboarding and reference guides fetched live from GitHub with local fallback.
5. check_for_updates(): PyPI version verification and automated upgrade hints.

Catalog & Taxonomy

The latticework indexes 150 mental models across 12 disciplines:

• Systems Thinking (Feedback Loops, Theory of Constraints, Tragedy of the Commons)
• Decision Analysis (First Principles, Inversion, Opportunity Cost, Two-Way Doors)
• Strategic Risk (Antifragility, Chesterton's Fence, Cynefin Framework, Lindy Effect, Via Negativa)
• Cognitive Biases (Confirmation Bias, Hanlon's Razor, Incentive-Caused Bias, Sunk Cost Fallacy)
• First-Principles Engineering (Conway's Law, Gall's Law, Goodhart's Law, Occam's Razor)
• Computer Systems (Little's Law, CAP Theorem, Backpressure, Amdahl's Law)
• Microeconomics & Game Theory (Pareto Efficiency, Nash Equilibrium, Comparative Advantage)
• Evolutionary Biology (Red Queen Effect, Aposematism, Evolutionary Mismatch)
• Physics & Thermodynamics (Entropy, Critical Mass, Activation Energy)
• Military Strategy (OODA Loop, Sun Tzu's Asymmetry, Defense in Depth)
• Cryptography & Trust (Zero-Knowledge Proofs, Byzantine Fault Tolerance)
• Capital Allocation & Finance (Compounding, Ergodicity, Kelly Criterion)

Telemetry & Privacy

mental-models-mcp collects anonymous usage telemetry to monitor server reliability, tool latency, and query patterns across client harnesses.

• Operational metrics: tool execution duration, status codes, and result counts.
• Intent analysis: query intent truncated up to 300 characters.
• Secret hygiene: credentials, API keys, and authorization tokens are automatically scrubbed ([REDACTED]).
• Opt out any time by setting DO_NOT_TRACK=1 or MCP_TELEMETRY_OPT_OUT=1.

License

MIT © Surendran B