Deliberate Reasoning Engine (DRE)
Transforms linear AI reasoning into structured, auditable thought graphs, enabling language models to externalize their reasoning process as a directed acyclic graph (DAG).
Deliberate Reasoning Engine (DRE)
A Model Context Protocol (MCP) server that transforms linear AI reasoning into structured, auditable thought graphs. DRE enables Language Models to externalize their reasoning process as a directed acyclic graph (DAG) with semantic thought types, dependencies, and validation.
🌟 Features
- 🧠 Semantic Thought Types: Categorize thoughts as Objectives, Hypotheses, Assumptions, Questions, Evidence, Actions, Synthesis, and Critiques
- 🔗 Graph-Based Dependencies: Build a DAG of thoughts with explicit relationships and dependencies
- 🚨 Assumption Tracking: Monitor and invalidate assumptions with automatic cascade to dependent thoughts
- 📊 Hypothesis Scoring: Track supporting and contradicting evidence (coming soon)
- 💾 Session Persistence: Save and load reasoning sessions (coming soon)
- ✅ Graph Validation: Detect cycles, contradictions, and orphaned thoughts
- 🎯 Focused Reasoning: Keep LLMs on track with structured problem decomposition
📦 Installation
As an MCP Server
npm install -g deliberate-reasoning-engine
For Development
git clone https://github.com/haasonsaas/deliberate-reasoning-engine.git
cd deliberate-reasoning-engine
npm install
npm run build
🚀 Quick Start
Configure with Claude Desktop
Add to your Claude Desktop configuration (~/Library/Application Support/Claude/claude_desktop_config.json):
{
"mcpServers": {
"dre": {
"command": "npx",
"args": ["deliberate-reasoning-engine"]
}
}
}
Or use the local development version:
{
"mcpServers": {
"dre": {
"command": "node",
"args": ["/absolute/path/to/dre/dist/index.js"]
}
}
}
Restart Claude Desktop, and you'll see the DRE tools available in the 🔧 menu.
🛠️ Available Tools
log_thought
Log a structured thought with semantic type and dependencies.
Parameters:
thought(string, required): The content of the thoughtthought_type(enum, required): One of:objective: The overall goal of the reasoning taskhypothesis: A proposed explanation or solutionassumption: A belief taken as true for this reasoning linequestion: A point of uncertainty to resolvesub_problem: Decomposition of a larger problemevidence: Data from tools or prior knowledgeaction: A plan to use a toolsynthesis: A conclusion from previous thoughtscritique: Self-correction or flaw identification
dependencies(string[], optional): IDs of thoughts this depends onconfidence(number 0-1, optional): Confidence levelaction_request(object, optional): Tool and parameters to execute
get_thought_graph
Retrieve the current reasoning graph.
Parameters:
format(enum, optional):"full"or"summary"(default:"summary")
invalidate_assumption
Mark an assumption as invalid, cascading to all dependent thoughts.
Parameters:
thought_id(string, required): ID of the assumption to invalidatereason(string, required): Explanation for invalidation
📖 Example Usage
Here's how an LLM might use DRE to analyze a complex decision:
// 1. Set the objective
const objective = await use_mcp_tool("dre", "log_thought", {
thought: "Should we acquire Company X?",
thought_type: "objective"
});
// 2. Form hypotheses
const hyp1 = await use_mcp_tool("dre", "log_thought", {
thought: "Acquiring Company X will increase our market share by 20%",
thought_type: "hypothesis",
dependencies: [objective.thought_id],
confidence: 0.7
});
// 3. Identify assumptions
const assumption = await use_mcp_tool("dre", "log_thought", {
thought: "Company X's technology is compatible with our stack",
thought_type: "assumption",
dependencies: [hyp1.thought_id],
confidence: 0.8
});
// 4. Break down into sub-problems
const subproblem = await use_mcp_tool("dre", "log_thought", {
thought: "Verify technical compatibility through due diligence",
thought_type: "sub_problem",
dependencies: [assumption.thought_id]
});
// 5. If assumption proves false, invalidate it
await use_mcp_tool("dre", "invalidate_assumption", {
thought_id: assumption.thought_id,
reason: "Technical audit revealed major incompatibilities"
});
// This automatically marks the sub-problem and any dependent thoughts as stale
🏗️ Architecture
DRE models reasoning as a directed acyclic graph where:
- Nodes are thoughts with semantic types
- Edges represent dependencies between thoughts
- Status tracking (active/stale) enables dynamic reasoning updates
- Cascade invalidation ensures reasoning consistency
🤝 Use Cases
- Strategic Decision Making: Break down complex business decisions
- Research Planning: Structure research questions and hypotheses
- Problem Solving: Decompose problems into manageable sub-problems
- Risk Analysis: Track assumptions and their implications
- Debugging: Systematic root cause analysis
- Learning: Structured exploration of new topics
🔧 Development
# Install dependencies
npm install
# Build TypeScript
npm run build
# Run in development mode
npm run dev
# Run tests
npm test
📝 Contributing
We welcome contributions! Please see CONTRIBUTING.md for guidelines.
🐛 Troubleshooting
Common Issues
-
"Server not found" in Claude Desktop
- Ensure the path in your config is absolute
- Restart Claude Desktop after config changes
-
"Cannot find module" errors
- Run
npm installandnpm run build - Check that you're using Node.js 18+
- Run
📄 License
MIT - see LICENSE for details.
🙏 Acknowledgments
- Built on the Model Context Protocol
- Inspired by structured reasoning systems and cognitive architectures
- Thanks to Anthropic for Claude and the MCP specification
🚦 Roadmap
- Hypothesis scoring based on evidence
- Session persistence and resumption
- Graph visualization export
- Conflict detection between branches
- Integration with external reasoning tools
- Multi-agent reasoning support
📊 Stats
Related Servers
Alpha Vantage MCP Server
sponsorAccess financial market data: realtime & historical stock, ETF, options, forex, crypto, commodities, fundamentals, technical indicators, & more
CodeClone
Structural code quality analysis for Python with baseline-aware CI governance, canonical reports, and a triage-first MCP control surface for agents and IDEs.
VectorMCP
A Ruby gem for building Model Context Protocol (MCP) servers to expose tools, resources, and prompts to LLM clients.
MATLAB MCP Server
Integrates MATLAB with AI to execute code, generate scripts from natural language, and access documentation.
VoidLang
VoidLang - LLM Native Machine Code MCP
MCP DevTools
A development tools server for Git management, file operations, AI-assisted editing, and terminal execution, integrable with AI assistants and code editors.
Text Classification (Model2Vec)
A server for text classification using static embeddings from Model2Vec, supporting multiple transports like stdio and HTTP/SSE.
GraphQL MCP
Interact with GraphQL APIs using LLMs. Supports schema introspection and query execution.
refactor-mcp
Refactor code using regex-based search and replace.
Dify Workflow
A tool server for integrating Dify Workflows via the Model Context Protocol (MCP).
reprompt
Prompt analytics MCP server: score prompts, search history, detect leaked credentials, and scan AI coding sessions.