Deep Thinking Engine

A collection of local MCP-style agents for cognitive enhancement, deep thinking, and systematic reasoning.

Overview

This project implements multiple specialized MCP (Model Context Protocol) agents that work together to enhance human cognitive capabilities:

🧠 Deep Thinking Engine

A comprehensive framework for breaking cognitive limitations through:

• Problem Decomposition: Break complex questions into manageable sub-problems
• Evidence Gathering: Leverage LLM web search for multi-source evidence collection
• Multi-Agent Debate: Organize structured debates from multiple perspectives
• Critical Evaluation: Apply Paul-Elder standards for rigorous thinking assessment
• Bias Detection: Identify and mitigate cognitive biases systematically
• Innovation Methods: Use SCAMPER/TRIZ for breakthrough thinking
• Socratic Reflection: Guide metacognitive awareness and self-assessment

Project Structure

mcp-style-agent/
├── .kiro/                     # Kiro IDE specs and configurations
│   └── specs/                 # Feature specifications
├── src/                       # Source code
│   └── mcps/                  # MCP collection
│       ├── deep_thinking/     # Deep thinking engine
│       └── shared/            # Shared utilities
├── tests/                     # Test suites
└── docs/                      # Documentation

Key Features

• 🔒 Privacy-First: Core reasoning runs locally, only search queries sent externally
• 🔧 Pluggable Architecture: YAML-configurable thinking flows and custom agents
• 📊 Transparent Process: Complete thinking traces with visualization
• 🎯 Scientific Methods: Based on cognitive science and learning research
• ⚡ Optimized Performance: Intelligent caching and async processing
• 🏗️ Modular Design: Shared components across multiple MCP agents

Quick Start

# Install with uv
uv sync

# Initialize the system
uv run deep-thinking init

# Start a thinking session
uv run deep-thinking think "How can we solve climate change effectively?"

MCP Server Deployment

The Deep Thinking Engine can be deployed as an MCP server for integration with MCP-compatible hosts like Cursor and Claude Desktop.

Using uvx (Recommended)

{
  "mcpServers": {
    "deep-thinking-engine": {
      "command": "uvx",
      "args": ["--from", "/path/to/mcp-style-agent", "deep-thinking-mcp-server"],
      "env": {
        "LOG_LEVEL": "INFO"
      }
    }
  }
}

Test Deployment

# Test uvx deployment
make test-uvx

# Start MCP server locally
make mcp-server

# Validate configuration
make mcp-server-validate

For detailed deployment instructions, see docs/deployment/README.md.

Development

# Setup development environment
uv sync --dev

# Run tests
uv run pytest

# Format code
uv run black .
uv run isort .

Architecture

The system uses a multi-agent architecture with specialized roles:

• Decomposer Agent: Question analysis and breakdown
• Evidence Seeker: Multi-source information gathering
• Debate Orchestrator: Structured multi-perspective analysis
• Critic Agent: Paul-Elder standards evaluation
• Bias Buster: Cognitive bias detection and mitigation
• Innovator Agent: SCAMPER/TRIZ creative thinking
• Reflector Agent: Socratic questioning and metacognition

License

MIT License - see LICENSE file for details.