Metacognitive Monitoring MCP Server

A systematic framework for self-monitoring knowledge boundaries, claim certainty, and reasoning quality to enhance metacognitive awareness and calibrated confidence.

Core Concepts

Knowledge Assessment

Knowledge assessments evaluate understanding within specific domains. Each assessment includes:

• Domain identification and scope
• Self-assessed knowledge level (expert to none)
• Confidence calibration (0.0-1.0)
• Supporting evidence for the assessment
• Known limitations and gaps
• Relevant training data cutoffs

Example:

{
  "domain": "Machine Learning Optimization",
  "knowledgeLevel": "proficient",
  "confidenceScore": 0.75,
  "supportingEvidence": "Familiar with gradient descent, Adam optimizer, and regularization techniques",
  "knownLimitations": ["Limited experience with advanced meta-learning algorithms"],
  "relevantTrainingCutoff": "2021-09"
}

Claim Assessment

Claim assessments classify and evaluate specific statements. They include:

• Statement classification (fact, inference, speculation, uncertain)
• Confidence scoring for the claim
• Evidence basis supporting the claim
• Alternative interpretations
• Falsifiability criteria

Example:

{
  "claim": "Transformer models require quadratic memory with sequence length",
  "status": "fact",
  "confidenceScore": 0.9,
  "evidenceBasis": "Self-attention mechanism computes all pairwise token interactions",
  "alternativeInterpretations": ["Linear attention variants exist but with trade-offs"],
  "falsifiabilityCriteria": "Discovery of attention mechanism with linear complexity and equivalent performance"
}

Reasoning Assessment

Reasoning assessments evaluate individual reasoning steps. They contain:

• Description of the reasoning step
• Potential cognitive biases
• Underlying assumptions
• Logical validity scoring (0.0-1.0)
• Inference strength evaluation (0.0-1.0)

Example:

{
  "step": "Since the model performs well on training data, it will generalize to new data",
  "potentialBiases": ["Confirmation bias", "Overfitting neglect"],
  "assumptions": ["Training data is representative", "Model complexity is appropriate"],
  "logicalValidity": 0.3,
  "inferenceStrength": 0.4
}

API

Tools

• metacognitiveMonitoring
  • Systematic self-monitoring of knowledge and reasoning quality
  • Input: Comprehensive metacognitive monitoring data structure
    • task (string): The task or question being addressed
    • stage (enum): Current monitoring stage - "knowledge-assessment" | "planning" | "execution" | "monitoring" | "evaluation" | "reflection"
    • knowledgeAssessment (object, optional): Domain knowledge evaluation
      • domain (string): Knowledge domain being assessed
      • knowledgeLevel (enum): "expert" | "proficient" | "familiar" | "basic" | "minimal" | "none"
      • confidenceScore (number): Confidence in assessment (0.0-1.0)
      • supportingEvidence (string): Evidence for knowledge level claim
      • knownLimitations (string[]): Known knowledge gaps
      • relevantTrainingCutoff (string, optional): Training data cutoff date
    • claims (array, optional): Specific claim assessments
      • claim (string): Statement being assessed
      • status (enum): "fact" | "inference" | "speculation" | "uncertain"
      • confidenceScore (number): Confidence in claim (0.0-1.0)
      • evidenceBasis (string): Supporting evidence
      • alternativeInterpretations (string[], optional): Alternative explanations
      • falsifiabilityCriteria (string, optional): Criteria for falsification
    • reasoningSteps (array, optional): Reasoning step evaluations
      • step (string): Description of reasoning step
      • potentialBiases (string[]): Identified cognitive biases
      • assumptions (string[]): Underlying assumptions
      • logicalValidity (number): Logical validity score (0.0-1.0)
      • inferenceStrength (number): Inference strength score (0.0-1.0)
    • overallConfidence (number): Overall confidence in conclusions (0.0-1.0)
    • uncertaintyAreas (string[]): Areas of significant uncertainty
    • recommendedApproach (string): Recommended approach based on assessment
    • monitoringId (string): Unique identifier for monitoring session
    • iteration (number): Current iteration of monitoring process
    • nextAssessmentNeeded (boolean): Whether further assessment is required
    • suggestedAssessments (array, optional): Suggested next assessments - "knowledge" | "claim" | "reasoning" | "overall"
  • Returns structured metacognitive analysis with visual confidence indicators
  • Supports iterative refinement of self-awareness and calibration
  • Tracks knowledge boundaries and reasoning quality over time

Code Mode Usage

This server supports Code Mode, allowing LLMs to import and use the functionality directly as a TypeScript API.

import { metacognitive } from "@wemake.cx/metacognitive-monitoring";

// Direct API usage
const result = await metacognitive.monitor({
  task: "Analyze complex system",
  stage: "planning",
  overallConfidence: 0.8,
  uncertaintyAreas: ["Edge cases"],
  recommendedApproach: "Systematic decomposition",
  monitoringId: "mon-1",
  iteration: 1,
  nextAssessmentNeeded: true
});

console.log(result.task); // "Analyze complex system"

Setup

Cursor

Add the following to your ~/.cursor/mcp.json file:

{
  "mcpServers": {
    "Metacognitive Monitoring": {
      "command": "bunx",
      "args": ["@wemake.cx/metacognitive-monitoring@latest"]
    }
  }
}

Raycast

Use Manage MCP Servers, press CMD + N and paste the following:

{
  "mcpServers": {
    "Metacognitive Monitoring": {
      "command": "bunx",
      "args": ["@wemake.cx/metacognitive-monitoring@latest"]
    }
  }
}

System Prompt

The prompt for utilizing metacognitive monitoring should encourage systematic self-assessment:

Follow these steps for metacognitive monitoring:

1. Knowledge Boundary Assessment:
   - Explicitly assess your knowledge level in the relevant domain
   - Identify specific areas of strength and limitation
   - Calibrate confidence based on evidence and experience
   - Acknowledge training data cutoffs and their implications

2. Claim Classification:
   - Distinguish between facts, inferences, speculation, and uncertainty
   - Provide evidence basis for each significant claim
   - Consider alternative interpretations of evidence
   - Establish falsifiability criteria where appropriate

3. Reasoning Quality Monitoring:
   - Evaluate each reasoning step for logical validity
   - Identify potential cognitive biases affecting judgment
   - Make underlying assumptions explicit
   - Assess inference strength and confidence

4. Uncertainty Management:
   - Identify areas of significant uncertainty
   - Recommend approaches based on confidence levels
   - Suggest additional assessments when needed
   - Iterate on understanding as new information emerges

5. Calibration and Iteration:
   - Track confidence calibration over time
   - Refine assessments based on feedback
   - Maintain awareness of knowledge boundaries
   - Continuously improve metacognitive accuracy

Usage Examples

Technical Domain Assessment

When working in specialized technical domains, systematically assess knowledge boundaries and claim confidence levels.

Complex Reasoning Chains

For multi-step reasoning, evaluate each step for biases, assumptions, and logical validity.

Uncertain Scenarios

In high-uncertainty situations, explicitly track confidence levels and identify areas requiring additional information.

Evidence Evaluation

When evaluating evidence, distinguish between different types of claims and their evidential basis.