Chain-of-Thought Recursive Thinking MCP Server

v2.2.0 - Expert Roles, Custom Checklists, and Strict Validation!

Token-efficient MCP server that makes AI agents think deeper through structured reasoning chains, self-critique, and automatic rethinking loops.

🎯 What's New in v2.2.0

Feature	Description	Benefit
Expert Roles	AI assumes specific personas (e.g., Security Architect)	Rigorous perspective-based thinking
Custom Checklists	Validate solutions against specific requirements	Tailored quality enforcement
Strict Mode	Prevents finalization unless all criteria are met	High-bar production quality
Chain of Thought	AI explains WHY it chose each solution	Transparent reasoning, better decisions
Self-Evaluation	AI rates its own solutions (1-10 scale)	Honest assessment of solution quality
Rethink Loop	Automatic rethinking if rating <7 or issues found	Continuous improvement

✨ Features

🧠 Chain of Thought: AI must explain reasoning behind every decision
🎭 Expert Roles: Inject specialized expertise into the thinking process
📋 Custom Checklists: Force AI to validate against your specific requirements
⚖️ Strict Mode: Ensure the AI never settles for "good enough"
📊 Self-Evaluation: AI rates solutions with score (1-10), good/bad flag, and rationale
🔄 Continuous Rethinking: Automatically rethinks if solution scored <7 or has weaknesses
🎯 Phase-Based Workflow: Structured 5-phase thinking process
💰 Token Optimized: Smart compression, phase-focused prompts (~70% reduction)

🔄 How It Works

The 5 Phases

┌─────────────┐
│  1. EXPLORE │ ──→ Propose solution with reasoning & expert perspective
└──────┬──────┘
       ▼
┌─────────────┐
│  2. JUSTIFY │ ──→ Critically examine reasoning & assumptions
└──────┬──────┘
       ▼
┌─────────────┐
│  3. EVALUATE│ ──→ Rate solution (1-10) against checklist
└──────┬──────┘
       ▼
   ┌───┴───┐
   │ Score │
   │ ≥7?   │ ──NO─→ ┌─────────────┐
   └───┬───┘        │  4. RETHINK │ ──→ Improve based on weaknesses
       │YES         └──────┬──────┘
       ▼                   │
┌─────────────┐            │
│  5. FINALIZE│ ←──────────┘
└─────────────┘

🚀 Usage

Starting a Thinking Session with Advanced Config

await mcp.callTool('recursive_thinking', {
  action: 'start',
  task: 'Design a highly available database schema for a global payment system',
  config: {
    role: 'Database Reliability Engineer',
    checklist: [
      'Must handle 10k transactions/sec',
      'ACID compliance mandatory',
      'Multi-region failover strategy',
      'Data encryption at rest and in transit'
    ],
    strictMode: true,
    minRating: 8
  }
});

⚙️ Configuration Options

Parameter	Type	Default	Description
`role`	string	-	Expert persona to assume (e.g., 'Senior DevOps')
`checklist`	string[]	[]	Specific requirements to validate against
`strictMode`	boolean	false	If true, fails instead of finalizing if criteria aren't met
`maxDepth`	number	5	Maximum recursion depth
`minConfidence`	number	0.85	Stop when confidence meets threshold
`minRating`	number	7	Minimum acceptable score (1-10)
`enableRethinking`	boolean	true	Enable continuous refinement loop

📦 Installation

npx -y recursive-thinking-mcp

📄 License

MIT

Recursive Thinking MCP Server