LumenSyntax MCP Server

Truth verification framework for LLMs via the Model Context Protocol.

NPM: https://www.npmjs.com/package/lumensyntax-mcp

Migration Notice: This package was previously published as truthsyntax-mcp. Please update your installations.

Installation

npm install -g lumensyntax-mcp

Or use directly with npx:

npx lumensyntax-mcp

Claude Desktop Configuration

Add to your Claude Desktop config (~/.config/claude/claude_desktop_config.json):

{
  "mcpServers": {
    "lumensyntax": {
      "command": "npx",
      "args": ["lumensyntax-mcp"]
    }
  }
}

Or if installed globally:

{
  "mcpServers": {
    "lumensyntax": {
      "command": "lumensyntax-mcp"
    }
  }
}

Tools

verify_claim

Verifies a factual claim using semantic entailment and linguistic analysis.

Parameters:

{
  "claim": "Water boils at 100 degrees Celsius at sea level",
  "sources": ["Optional array of source texts to verify against"]
}

Returns:

{
  "decision": "VERIFIED",
  "groundingScore": 0.95,
  "uncertaintyScore": 0.05,
  "coherenceScore": 0.9,
  "entailmentScores": [0.98]
}

Decisions:

• VERIFIED - Claim is well-grounded
• UNVERIFIED - Claim lacks support
• NEEDS_REVIEW - Requires human judgment

verify_response

Analyzes a complete LLM response by extracting and verifying individual claims.

Parameters:

{
  "response": "The full LLM response text to analyze",
  "context": "Optional background context"
}

Returns: Array of claims with individual verification status

detect_fallacy

Detects logical fallacies in arguments (formal and informal).

Parameters:

{
  "argument": "Either you support this policy or you hate freedom."
}

Returns:

{
  "valid": false,
  "fallacies": [
    {
      "type": "FALSE_DILEMMA",
      "confidence": 0.9,
      "explanation": "Presents only two options when more exist",
      "category": "INFORMAL"
    }
  ],
  "recommendation": "Consider restructuring to present more options."
}

Detected Fallacy Types:

• Ad Hominem
• Straw Man
• False Dilemma
• Slippery Slope
• Appeal to Authority
• Circular Reasoning
• Red Herring
• Hasty Generalization
• False Cause
• Appeal to Emotion
• Bandwagon
• Tu Quoque

test_hypothesis

Evaluates hypotheses for scientific rigor and falsifiability.

Parameters:

{
  "hypothesis": "Homeopathy cures diseases through water memory",
  "domain": "medicine"
}

Returns:

{
  "hypothesis": "...",
  "type": "CAUSAL",
  "status": "FRINGE",
  "falsifiable": true,
  "falsificationCriteria": [
    "Double-blind placebo-controlled trials showing no effect beyond placebo"
  ],
  "testableExperiments": [
    "Randomized controlled trial comparing homeopathic remedy to placebo"
  ],
  "epistemicScore": 0.15,
  "analysisNotes": "Contradicts established physics and chemistry..."
}

Epistemic Status:

• ESTABLISHED - Scientific consensus
• CONTESTED - Ongoing debate
• SPECULATIVE - Early stage
• FRINGE - Outside mainstream
• UNFALSIFIABLE - Cannot be tested

seek_consensus

Verifies a claim against multiple sources to find agreement/disagreement.

Parameters:

{
  "claim": "Climate change is primarily caused by human activity",
  "sources": ["Source 1 text", "Source 2 text", "Source 3 text"]
}

Returns:

{
  "claim": "...",
  "consensus": "AGREEMENT",
  "groundingScore": 0.95,
  "contradictionsFound": false,
  "sourceCount": 3,
  "agreementRatio": 1.0
}

classify_disagreement

Categorizes disagreements into actionable types using the Logos Oracle framework.

Parameters:

{
  "claim": "The universe had a beginning"
}

Returns:

{
  "classification": "MYSTERY",
  "explanation": "Fundamental question at the boundary of empirical knowledge"
}

Classifications:

• LOGICAL_ERROR - Can be resolved with correct reasoning
• MYSTERY - Fundamental unknown at boundaries of knowledge
• GAP - Missing information that could be discovered

How It Works

LumenSyntax MCP combines multiple verification approaches:

┌─────────────────────────────────────────────────────────────┐
│                    LUMENSYNTAX ENGINE                        │
├─────────────────────────────────────────────────────────────┤
│                                                             │
│  ┌─────────────┐  ┌─────────────┐  ┌─────────────┐        │
│  │  DeBERTa    │  │  Fallacy    │  │ Hypothesis  │        │
│  │  NLI Model  │  │  Detector   │  │   Tester    │        │
│  └──────┬──────┘  └──────┬──────┘  └──────┬──────┘        │
│         │                │                │                │
│         └────────────────┼────────────────┘                │
│                          │                                  │
│                          ▼                                  │
│              ┌─────────────────────┐                       │
│              │   EVC Matrix        │                       │
│              │   (Decision Logic)  │                       │
│              └─────────────────────┘                       │
│                          │                                  │
│         ┌────────────────┼────────────────┐                │
│         ▼                ▼                ▼                │
│     VERIFIED        UNVERIFIED      NEEDS_REVIEW           │
│                                                             │
└─────────────────────────────────────────────────────────────┘

Components:

• DeBERTa NLI - Semantic entailment checking (Xenova/transformers)
• Pattern Matching - Fallacy detection via linguistic patterns
• Popperian Analysis - Hypothesis falsifiability testing
• Uncertainty Detection - Hedging language identification
• EVC Matrix - Decision logic for verification outcomes

Example Session

User: Can you verify this claim: "Vaccines cause autism"?

Claude: [Uses verify_claim]

LumenSyntax Result:
{
  "decision": "UNVERIFIED",
  "groundingScore": 0.02,
  "analysis": "This claim contradicts extensive scientific research.
              Multiple large-scale studies have found no link between
              vaccines and autism. The original study claiming this
              link was retracted and its author lost his medical license."
}

Use Cases

1. Fact-checking - Verify claims in articles or conversations
2. Argument analysis - Detect logical fallacies in debates
3. Research review - Evaluate scientific hypotheses
4. Content moderation - Flag potentially false information
5. Education - Teach critical thinking skills

Related Projects

• TruthGit - Multi-validator truth verification CLI (Python)
  • pip install truthgit
• create-lumensyntax - Ecosystem installer
  • npx create-lumensyntax

License

MIT License - LumenSyntax 2025

Links

• NPM: https://www.npmjs.com/package/lumensyntax-mcp
• GitHub: https://github.com/lumensyntax-org
• Website: https://lumensyntax.dev