deep-thinker

Advanced cognitive thinking MCP server with DAG-based thought graph, 10 reasoning strategies (including auto-selection), 17 tools, node aliases, session persistence, structured responses, and intelligent error handling.

A significant evolution beyond sequential-thinking MCP, providing structured deep reasoning with graph-based thought management, schema validation, and intelligent strategy selection.

Quick Start

npx deep-thinker

{
  "mcpServers": {
    "deep-thinker": {
      "command": "npx",
      "args": ["-y", "deep-thinker"]
    }
  }
}

Examples

Features

• DAG-Based Thought Graph — Thoughts form a directed acyclic graph with branching, merging, and cross-edges (not just a linear chain)
• 10 Reasoning Strategies — Sequential, Dialectic (thesis→antithesis→synthesis), Parallel, Analogical, Abductive, First Principles (deconstruct to fundamentals), Counterfactual (what-if with ripple effects), Systems Thinking (feedback loops & leverage points), MCTS (Monte Carlo optimization), Auto (intelligent auto-selection based on content and graph state)
• Node Aliases — Use "last", "best", "root" instead of cryptic node IDs for any nodeId parameter
• Structured Responses — All tool responses return consistent MCPResponse JSON with status, summary, confidence, nextSuggested action
• Session Persistence — Auto-saves thought graph to ~/.deep-thinker/sessions/; resume across MCP restarts with reset({ resume: "name" })
• Friendly Error Messages — Zod validation errors translated to human-readable hints (e.g., "confidence 0 ile 1 arasında...")
• Confidence Scoring — Multi-factor confidence evaluation with support/contradiction analysis, depth penalties, and knowledge integration boosts
• Self-Critique — Automatic critique generation with severity levels and confidence adjustments
• Metacognitive Engine — Detects stuck states, stagnation, declining confidence; suggests strategy switches and corrective actions
• Knowledge Integration — Attach external knowledge to thoughts, detect gaps, validate consistency across sources
• Thought Pruning — Dead-end detection, redundancy removal, deep unproductive branch elimination, path optimization
• help Tool — Discover all 17 tools grouped by category (core/advanced/workflow) with quick-start examples
• conclude Tool — Comprehensive graph summary with primaryFinding, actionItems, graphHealth, and nextSuggested
• High-IQ Reasoning Enhancements — 8 advanced tools: visualization, devil's advocate, cross-disciplinary synthesis, temporal projection, ethical evaluation, emotional intelligence analysis, decision explanation, social impact analysis
• Emotional Intelligence — Analyze emotional tone, empathy, persuasion effectiveness, stakeholder emotions
• Ethical Frameworks — Evaluate through deontological, consequentialist, virtue ethics, rights-based perspectives
• Cross-Domain Synthesis — Combine insights from biology, economics, physics, psychology, computer science, art
• Temporal Reasoning — Project thoughts into future/past scenarios with optimistic, pessimistic, realistic, disruptive scenarios
• Social Impact Modeling — Analyze stakeholder emotions, group cohesion, persuasion effectiveness, ethical alignment
• Uncertainty Quantification — Confidence intervals, probability distributions, sensitivity analysis for robust decisions
• Multi-Language Support — Thoughts in English, Turkish, German, French, Spanish, Japanese, Chinese, Russian
• Meta-Cognitive Layers — Recursive reasoning across 5 levels of meta-cognition
• PromptOptimizer (Node Zero) — Entry point that transforms vague prompts into optimized Super Prompts with automatic strategy routing

Installation

Global

npm install -g deep-thinker

npx (no install)

npx deep-thinker

MCP Configuration

Claude Desktop

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "deep-thinker": {
      "command": "npx",
      "args": ["-y", "deep-thinker"]
    }
  }
}

Or if installed globally:

{
  "mcpServers": {
    "deep-thinker": {
      "command": "deep-thinker"
    }
  }
}

Other MCP Clients

The server communicates over stdio. Point your MCP client to the deep-thinker command or node path/to/dist/index.js.

Response Format

All tool responses follow the MCPResponse structure:

{
  "status": "ok | error | warning",
  "nodeId": "thought_3",
  "summary": "sequential stratejisiyle \"Should we use microservices?...\" eklendi",
  "confidence": 0.75,
  "data": { "...": "tool-specific data" },
  "nextSuggested": {
    "tool": "evaluate",
    "params": { "critique": true },
    "reason": "Düşük confidence — değerlendirme önerilir"
  },
  "warnings": ["Stuck detected: ..."]
}

The nextSuggested field always recommends the next logical step, making it easy to chain tool calls without guessing.

Node Aliases

Instead of looking up cryptic node IDs, use aliases for any nodeId, parentId, or targetId parameter:

evaluate({ nodeId: "last" })                              → evaluates the latest thought
simulate_devils_advocate({ nodeId: "best", depth: 2 })    → challenges the strongest thought
graph({ action: "path", nodeId: "root", targetId: "best" }) → traces from root to best conclusion

Session Persistence

Thought graphs are automatically saved after every think call. Sessions are stored in ~/.deep-thinker/sessions/.

// Save current session explicitly
reset({ save: true, saveName: "my-analysis" })

// List saved sessions
reset({ listSessions: true })

// Resume a saved session after MCP restart
reset({ resume: "my-analysis" })

Tools

Core Tools

think

Add a thought to the cognitive graph using a reasoning strategy.

Parameters:

Strategy details:

How auto strategy works:

The auto strategy analyzes your content for keywords and the current graph state:

• Content with "why"/"neden"/"how"/"nasıl" → abductive
• Content with "if"/"eğer"/"what if"/"varsayalım" → counterfactual
• Content with "vs"/"veya"/"compare"/"karşılaştır" → dialectic
• Content with "system"/"sistem"/"loop"/"döngü" → systems_thinking
• Content with "fundamental"/"temel"/"assumption"/"varsayım" → first_principles
• Low avg confidence + many nodes → parallel (break through impasse)
• First thought → sequential
• After 4+ sequential thoughts → dialectic (introduce opposing view)
• Default → sequential

Edge types: derives_from, contradicts, supports, refines, challenges, synthesizes, parallels, abstracts, instantiates

evaluate

Evaluate the thinking process with confidence scoring, critique, and graph health analysis.

Parameters:

metacog

Metacognitive operations — monitor and control the thinking process.

Parameters:

The metacognitive engine automatically:

• Detects stagnation (confidence not improving)
• Detects declining confidence trends
• Detects excessive contradictions
• Suggests strategy switches, pruning, backtracking, or concluding

graph

Query and visualize the thought graph.

Parameters:

prune

Prune and optimize the thought graph.

Parameters:

reset

Reset the thought graph and start a fresh session, save, or resume a saved session.

Parameters:

conclude

Analyze the entire thought graph and produce a comprehensive summary-conclusion with action items and graph health report.

Parameters:

Response includes:

• primaryFinding — Top conclusion with confidence
• supportingEvidence — Additional high-confidence nodes
• strategiesUsed — Which strategies contributed
• keyInsights — Insight-type nodes from the best path
• actionItems — Prioritized actions derived from conclusions
• graphHealth — Node count, dead ends, avg confidence, recommendation
• nextSuggested — Logical next step (prune if unhealthy, save if done)

help

Discover deep-thinker tools and learn usage workflows.

Parameters:

Categories:

• core — 7 daily-use tools (think, evaluate, metacog, graph, prune, reset, conclude)
• advanced — 8 deep-analysis tools (visualization, devil's advocate, cross-disciplinary, temporal, ethical, emotional, explanation, social impact, prompt optimizer)
• workflow — 3 recommended workflows:
  • Quick Decision — reset → think parallel → evaluate → conclude
  • Deep Analysis — reset → first_principles → counterfactual → devil's advocate → evaluate → metacog → prune → conclude
  • Breaking Dead Ends — metacog report → switch strategy → cross-disciplinary → abductive

Enhanced Tools (High-IQ Reasoning)

visualize_thought_graph

Generate visual representation of the thought graph as SVG or ASCII.

Parameters:

simulate_devils_advocate

Generate counterarguments and opposing viewpoints for a given thought.

Parameters:

cross_disciplinary_synthesis

Combine insights from multiple domains to generate novel perspectives.

Parameters:

temporal_projection

Project thoughts into future or past scenarios.

Parameters:

ethical_framework_evaluation

Evaluate a thought or decision through multiple ethical frameworks.

Parameters:

emotional_intelligence_analysis

Analyze emotional tone, stakeholder emotions, and social dynamics.

Parameters:

explain_decision

Generate human-understandable explanation of a decision path.

Parameters:

social_impact_analysis

Analyze social impact, stakeholder emotions, group cohesion, and persuasion effectiveness.

Parameters:

optimize_prompt

PromptOptimizer (Node Zero) — transform vague prompts into optimized Super Prompts with strategy routing.

Parameters:

Usage Examples

Auto Strategy Selection (New!)

think: { content: "Eğer mikroservis kullansaydık ne olurdu?" }
→ Auto-selects strategy: counterfactual (detected "Eğer" = "if" signal)

think: { content: "Why is the server crashing?" }
→ Auto-selects strategy: abductive (detected "why" signal)

think: { content: "Monolith vs microservices?" }
→ Auto-selects strategy: dialectic (detected "vs" comparison signal)

Sequential Reasoning

think: "Should we use microservices?" → type: question, confidence: 0.9
think: "Monolith has deployment bottlenecks" → type: analysis, confidence: 0.7
think: "Team lacks DevOps capacity for microservices" → type: evidence, confidence: 0.8
evaluate: { nodeId: "last", critique: true }
→ { status: "ok", confidence: 0.73, nextSuggested: { tool: "metacog" } }

Dialectic Reasoning

think: {
  strategy: "dialectic",
  dialectic: {
    thesis: "Microservices improve scalability",
    antithesis: "But add operational complexity",
    synthesis: "Use modular monolith as middle ground"
  },
  confidence: 0.75
}

Using Node Aliases

evaluate({ nodeId: "last" })                              → evaluate latest thought
simulate_devils_advocate({ nodeId: "best", depth: 3 })    → challenge strongest thought
graph({ action: "path", nodeId: "root", targetId: "best" }) → trace reasoning path
think({ parentId: "root", content: "Alternative..." })    → branch from root

Session Save & Resume

// Work on a problem...
think({ content: "Analysis...", strategy: "auto" })
think({ content: "Another insight..." })

// Save before closing
reset({ save: true, saveName: "architecture-review" })

// ... MCP restarts ...

// Resume exactly where you left off
reset({ resume: "architecture-review" })
→ { status: "ok", summary: "architecture-review oturumu geri yüklendi — 5 node ile devam ediliyor" }

First Principles Reasoning

think: {
  strategy: "first_principles",
  firstPrinciples: {
    problem: "How to improve battery efficiency?",
    assumptions: ["Batteries must use lithium", "Charging takes hours"],
    depth: 3,
    domain: "physics"
  }
}
→ Creates: Problem → Assumptions Challenged → Fundamental Truths → Reconstructed Solution

Counterfactual (What-If) Analysis

think: {
  strategy: "counterfactual",
  counterfactual: {
    currentState: "Office-based work with 5-day commute",
    variablesToChange: [
      { variable: "work_location", currentValue: "office", hypotheticalValue: "remote", impactWeight: 0.9 },
      { variable: "commute_days", currentValue: 5, hypotheticalValue: 0, impactWeight: 0.8 }
    ],
    timeHorizon: "medium_term",
    rippleDepth: 3
  }
}
→ Creates: Baseline → Variable Changes → Stage 1/2/3 Ripple Effects → Scenarios → Risk Analysis

Systems Thinking

think: {
  strategy: "systems_thinking",
  systemsThinking: {
    systemDescription: "Software development team dynamics",
    components: [
      { name: "FeatureBacklog", type: "stock", description: "Pending work" },
      { name: "DeveloperCapacity", type: "stock", description: "Available developers" },
      { name: "CodeReviews", type: "flow", description: "Review process" },
      { name: "Quality", type: "converter", description: "Quality gates" }
    ],
    focusArea: "feedback_loops"
  }
}
→ Creates: System Overview → Components → Feedback Loops → Leverage Points → Recommendations

MCTS (Monte Carlo Tree Search)

think: {
  strategy: "mcts",
  mcts: {
    problem: "Which architecture pattern to choose?",
    possibleActions: [
      { id: "microservices", description: "Microservices architecture", estimatedReward: 0.7 },
      { id: "monolith", description: "Monolithic architecture", estimatedReward: 0.5 },
      { id: "modular", description: "Modular monolith", estimatedReward: 0.8 }
    ],
    numSimulations: 100,
    pruningThreshold: 0.2
  }
}
→ Creates: Root → Actions → Simulations → Pruning Analysis → Optimal Path

Conclude Analysis

conclude({ detailLevel: "detailed" })
→ {
  status: "ok",
  summary: "12 dusunce, 3 dal, sequential+counterfactual stratejileriyle analiz tamamlandi",
  data: {
    conclusion: { primaryFinding: "...", confidence: 0.85 },
    actionItems: [{ action: "Investigate...", priority: "high" }, ...],
    graphHealth: { totalThoughts: 12, avgConfidence: 0.72, recommendation: "Graf saglikli gorunuyor" }
  },
  nextSuggested: { tool: "reset", params: { save: true }, reason: "Analizi kaydetmeyi unutmayin" }
}

Metacognitive Guidance

metacog: { action: "auto_update" }
→ Stuck detected + suggested action in nextSuggested

metacog: { action: "switch", strategy: "parallel", reason: "Break through impasse" }
→ Strategy switched + next step recommended

Pruning

prune: { action: "analyze" }
→ Dead Ends, Redundant Branches, Total prunable count

prune: { action: "prune" }
→ Nodes pruned + metacog updated + nextSuggested

Getting Help

help()                    → all tools, all categories, all workflows
help({ category: "core" })     → 7 core tools with quick-start examples
help({ category: "advanced" }) → 9 advanced tools
help({ category: "workflow" }) → 3 recommended workflows

Friendly Error Messages

When validation fails, you get human-readable errors instead of raw Zod output:

think({ confidence: 1.5 })
→ {
  status: "error",
  error: "VALIDATION_ERROR",
  message: "\"confidence\" parametresinde hata: ...",
  field: "confidence",
  hint: "confidence 0 ile 1 arasında bir sayı olmalı. Örnek: confidence: 0.7"
}

evaluate({ nodeId: "nonexistent" })
→ {
  status: "error",
  error: "NODE_NOT_FOUND",
  provided: "nonexistent",
  hint: "Geçerli alias'lar: \"last\", \"best\", \"root\" veya graph aracıyla node ID alın"
}

Architecture

src/
├── index.ts                         MCP server & 17 tool handlers
├── test.ts                          Core functionality tests (118 tests)
├── test_enhanced_strategies.ts      Strategy tests (13 tests)
├── core/
│   ├── types.ts                     Type definitions, MCPResponse, NextAction
│   ├── schemas.ts                   Zod validation schemas (10 strategies incl. auto)
│   ├── node.ts                      ThoughtNode CRUD operations
│   ├── graph.ts                     DAG-based thought graph + resolveNodeId + aliases
│   ├── strategies.ts               10 reasoning strategies + selectStrategy (auto)
│   ├── scorer.ts                    Confidence scoring & self-critique
│   ├── metacog.ts                   Metacognitive engine with smart triggers
│   ├── knowledge.ts                 Knowledge integration & validation
│   ├── pruner.ts                    Dead-end/redundancy detection & pruning
│   ├── session.ts                   Session persistence (save/load/resume)
│   └── errors.ts                    Friendly error formatting (Zod + unknown)
└── enhancements/
    ├── visualization.ts            SVG & ASCII graph visualization
    ├── devils_advocate.ts           Counterargument generation
    ├── cross_disciplinary.ts        Cross-domain analogy engine
    ├── temporal_projection.ts       Future/past thought projection
    ├── ethical_evaluation.ts         4 ethical frameworks
    ├── emotional_intelligence.ts    Emotion & sentiment analysis
    ├── explanation.ts               Decision explainability
    └── social_impact.ts             Stakeholder & social impact

What's New in v3.0.0

Comparison with sequential-thinking

Development

git clone https://github.com/hubinoretros/deep-thinker.git
cd deep-thinker
npm install
npm run build
npm start

Testing

npm run build
npm test

131 tests covering all modules: Node, Graph, 10 Strategies (including auto), Scorer, Metacog, Knowledge, Pruner, Integration, Edge Cases, Schema Validation.

Documentation

• Architecture Deep Dive — how the DAG, scoring, metacog, and pruning work internally
• Strategy Selection Guide — when to use each strategy and how to combine them

Contributing

See CONTRIBUTING.md for guidelines. PRs welcome — especially new reasoning strategies and MCP tool ideas.

License

MIT