Collaborative Reasoning MCP Server

A sophisticated MCP server for simulating expert collaboration with diverse perspectives to tackle complex problems through structured multi-persona reasoning.

Core Concepts

Expert Personas

The server creates and manages expert personas with distinct characteristics:

• Identity: Unique ID, name, and background
• Expertise: Specific areas of knowledge and experience
• Perspective: Unique viewpoint and approach to problems
• Biases: Acknowledged limitations and potential blind spots
• Communication Style: Tone and approach to interaction

Example persona:

{
  "id": "tech-lead",
  "name": "Sarah Chen",
  "expertise": ["software architecture", "scalability", "team leadership"],
  "background": "15 years in enterprise software development",
  "perspective": "Pragmatic focus on maintainable, scalable solutions",
  "biases": ["over-engineering tendency", "preference for proven technologies"],
  "communication": {
    "style": "analytical",
    "tone": "direct"
  }
}

Collaborative Process

The reasoning process follows structured stages:

1. Problem Definition: Clarify the challenge and scope
2. Ideation: Generate diverse ideas and approaches
3. Critique: Evaluate and challenge proposed solutions
4. Integration: Synthesize insights from different perspectives
5. Decision: Reach consensus or identify trade-offs
6. Reflection: Extract learnings and next steps

Contribution Types

Personas contribute through various interaction types:

• Observation: Factual insights or data points
• Question: Clarifying or probing inquiries
• Insight: Novel connections or understanding
• Concern: Potential risks or limitations
• Suggestion: Proposed solutions or approaches
• Challenge: Constructive disagreement or alternative view
• Synthesis: Integration of multiple perspectives

Disagreement Management

The system tracks and resolves disagreements through:

• Position Mapping: Clear articulation of different viewpoints
• Argument Tracking: Supporting evidence for each position
• Resolution Types: Consensus, compromise, integration, or tabling
• Productive Conflict: Leveraging disagreement for better outcomes

API

Code Mode Usage

This server supports the MCP Code Mode architecture, allowing direct programmatic usage in TypeScript applications.

Installation

npm install @wemake.cx/collaborative-reasoning

Example

import { CollaborativeReasoning } from "@wemake.cx/collaborative-reasoning";

const collaborativeReasoning = new CollaborativeReasoning();

const result = await collaborativeReasoning.collaborate({
  topic: "Should we use microservices?",
  personas: [
    {
      id: "tech-lead",
      name: "Sarah Chen",
      expertise: ["Architecture"],
      background: "Senior Architect",
      perspective: "Scalability focused",
      biases: [],
      communication: { style: "direct", tone: "formal" }
    }
  ],
  contributions: [],
  stage: "problem-definition",
  activePersonaId: "tech-lead",
  sessionId: "session-1",
  iteration: 1,
  nextContributionNeeded: true
});

console.log(`Next persona: ${result.nextPersonaId}`);

// Generate visualization
console.log(collaborativeReasoning.visualize(result));

MCP Tools

• collaborativeReasoning
  • Simulates expert collaboration with diverse perspectives
  • Input: Comprehensive collaboration data structure
    • topic (string): The problem or challenge being addressed
    • personas (array): Expert personas with expertise, background, and communication style
    • contributions (array): Contributions from personas with type, content, and confidence
    • disagreements (array, optional): Points of disagreement and their resolution
    • stage (enum): Current collaboration stage (problem-definition, ideation, critique, integration, decision, reflection)
    • activePersonaId (string): Currently active persona
    • nextPersonaId (string, optional): Next persona to contribute
    • keyInsights (array, optional): Key insights from the collaboration
    • consensusPoints (array, optional): Points of agreement
    • openQuestions (array, optional): Unresolved questions
    • finalRecommendation (string, optional): Final collaborative recommendation
    • sessionId (string): Unique session identifier
    • iteration (number): Current iteration number
    • nextContributionNeeded (boolean): Whether more input is needed
    • suggestedContributionTypes (array, optional): Suggested next contribution types
  • Output: Structured collaboration analysis with visual representation
    • Formatted display of personas, contributions, disagreements, and insights
    • Progress tracking and next steps
    • Confidence levels and consensus points
  • Validates all input data and maintains session history
  • Provides colorized visual output for better readability

Setup

bunx

{
  "mcpServers": {
    "Collaborative Reasoning": {
      "command": "bunx",
      "args": ["@wemake.cx/collaborative-reasoning@latest"]
    }
  }
}

bunx with custom settings

The server supports various configuration options:

{
  "mcpServers": {
    "Collaborative Reasoning": {
      "command": "bunx",
      "args": ["@wemake.cx/collaborative-reasoning@latest"],
      "env": {
        "COLLABORATION_MAX_PERSONAS": "8",
        "COLLABORATION_MAX_ITERATIONS": "20",
        "COLLABORATION_CONFIDENCE_THRESHOLD": "0.7",
        "COLLABORATION_VISUAL_OUTPUT": "true"
      }
    }
  }
}

• COLLABORATION_MAX_PERSONAS: Maximum number of personas per session (default: 8)
• COLLABORATION_MAX_ITERATIONS: Maximum collaboration iterations (default: 20)
• COLLABORATION_CONFIDENCE_THRESHOLD: Minimum confidence for consensus (default: 0.7)
• COLLABORATION_VISUAL_OUTPUT: Enable colorized visual output (default: true)

System Prompt

The prompt for utilizing collaborative reasoning should encourage diverse perspective integration:

Follow these steps for collaborative reasoning:

1. Problem Framing:
   - Define the challenge clearly and comprehensively
   - Identify key stakeholders and their interests
   - Establish success criteria and constraints
   - Set up diverse expert personas with complementary expertise

2. Multi-Perspective Analysis:
   - Engage each persona to contribute their unique viewpoint
   - Encourage different types of contributions (observations, insights, concerns)
   - Surface assumptions and biases explicitly
   - Promote constructive disagreement and debate

3. Synthesis and Integration:
   - Identify points of consensus and disagreement
   - Explore creative combinations of different approaches
   - Address concerns and limitations raised by personas
   - Build on insights through cross-pollination of ideas

4. Decision and Recommendation:
   - Evaluate trade-offs between different approaches
   - Seek win-win solutions that address multiple perspectives
   - Document remaining uncertainties and risks
   - Provide clear, actionable recommendations

5. Reflection and Learning:
   - Extract key insights and lessons learned
   - Identify areas for further exploration
   - Document the reasoning process for future reference
   - Plan next steps and follow-up actions