Hallucination Detector Prover MCP Connector for Claude
A+LLMs present fabricated information as fact. This tool forces epistemic rigor: cite verifiable sources for every claim, quantify confidence per assertion, separate facts from opinions, declare knowledge boundaries, and cross-reference for internal contradictions.
The most dangerous failure mode of LLMs is not producing wrong answers — it is producing wrong answers that sound right.
The 5 Grounding Failures This Tool Detects
- SOURCE_MISSING — Factual claims without verifiable citations. 'Studies show' is not a source.
- CONFIDENCE_UNCALIBRATED — Uncertainty hidden behind absolute language.
- OPINION_AS_FACT — Subjective judgments smuggled as objective data.
- KNOWLEDGE_EXCEEDED — Claims beyond training data or accessible sources.
- SELF_CONTRADICTING — Internal contradictions across paragraphs.
The engine enforces 5 Decision Pivots with 5 semantic trap lists that catch weasel citations, false certainty, smuggled opinions, fabricated access, and circular self-validation.
Related Connectors
Agent Hallucination Cross-Checker MCP
A deterministic engine to audit agent outputs by measuring consensus, source validity, and semantic contradictions.
Hallucination Detection Score MCP
Quantify the reliability of AI agent outputs using deterministic hallucination scoring.
Claude Sycophancy Detector MCP
Detects when AI models agree with incorrect user assumptions by verifying factual claims against codebase metadata.
Edison Experimentation Prover MCP
A team chose paper filing because 'best practice.' No pilot. No alternatives tested. 8 months later, 14,000 submissions/day — 47 hours behind on retrieval. Emergency migration under pressure. Cost: 3x the estimate, 6 weeks frozen. Edison tested 3,000+ filament materials before carbonized bamboo — he did not pick the 'obvious choice.' This tool forces experimentation: test alternatives with measured criteria, iterate beyond the first solution, build the ecosystem, prove viability under real conditions, and document dead ends.