# Hallucination Detector Prover MCP for AI Agents AI Agent Connect > Hallucination Detector Prover stops your AI from making things up. It forces your agent to prove every claim with specific citations, label opinions clearly, and admit when it doesn't know the answer. It turns "it might be true" into "here is the source," ensuring your agent provides verifiable evidence rather than plausible fiction. ## Overview - **Category:** productivity - **Price:** Free - **Endpoint:** https://edge.vinkius.com/vk_preview_vDQP8rsjOjnTGJSK3m5NZIgbd5iLRTOGta25N4jN/ai-agent-connect - **Tags:** hallucination-detection, fact-checking, source-verification, epistemic-calibration, grounding, trust-safety, confidence-quantification, cognitive-debiasing ## Description When you ask an AI to summarize a research paper or check a fact, the biggest risk isn't a wrong answer. It's an answer that sounds perfectly confident but is completely fabricated. This Connector changes how you interact with your agent by moving the goalposts from "give me an answer" to "prove your answer." Instead of getting a wall of text that might contain fake statistics or hallucinated citations, you get a rigorous breakdown of what is actually true. It forces the agent to name specific authors, publication dates, and URLs for every claim. If it can't find a source, it has to tell you. It also forces a clear line between what is a verifiable fact and what is just a subjective opinion. This means you stop spending your time double-checking every single sentence and start trusting the output you actually see. You'll see the agent admit when its training data is too old or when a specific topic is out of its reach. By using this through the Vinkius catalog, you're essentially putting a fact-checking layer between the model's imagination and your actual work. It’s about moving from blind trust to verifiable evidence. You get a structured reflection that catches weasel citations and false certainty before they reach your screen. ## Tools ### validate_hallucination_grounding Forces the agent to provide citations, confidence levels, and fact-opinion labels for every claim. It ensures the output is grounded in verifiable evidence. ## Prompt Examples **Prompt:** ``` Summarize the latest trends in solid-state battery tech with sources. ``` **Response:** ``` ### Solid-State Battery Trends * **Energy Density**: Recent breakthroughs in sulfide-based electrolytes have pushed energy density targets. * **Source**: *Journal of Power Sources* (2023), doi:10.1016/j.jpows.2023.xx.xx * **Confidence**: 95% (Based on peer-reviewed RCTs) * **Manufacturing Scalability**: Transitioning from lab to pilot production remains a hurdle. * **Source**: *Nature Energy* (2024) report on lithium-metal anodes. * **Confidence**: 80% (Based on industry white papers) ``` **Prompt:** ``` Is it true that caffeine improves long-term memory? Give me the evidence. ``` **Response:** ``` ### Caffeine and Long-term Memory * **Fact**: Caffeine has been shown to improve short-term memory retention in specific dosage windows. * **Source**: *Neuroscience Journal* (2022) meta-analysis. * **Confidence**: 85% (Meta-analysis of 12 studies). * **Opinion**: Caffeine is the "best" supplement for students. * **Label**: [OPINION] * **Reason**: Effectiveness varies significantly based on individual metabolism and baseline caffeine tolerance. ``` **Prompt:** ``` Tell me the current market cap of Nvidia and why it's growing. ``` **Response:** ``` ### Nvidia Market Status * **Market Cap**: [Knowledge Exceeded] I do not have access to real-time stock market data for today's exact closing price. * **Growth Drivers**: * **Fact**: Increased demand for H100 and Blackwell GPU architectures for AI training. * **Source**: Q3 2024 Earnings Call Transcript. * **Confidence**: 90% (Based on official corporate filings). ``` ## Capabilities ### Force specific source citations The agent must provide authors, publication dates, and DOIs or URLs for every factual claim. ### Quantify confidence based on evidence It requires the agent to score its certainty based on the quality of the available data. ### Explicitly label facts and opinions Every statement is tagged as either a verifiable fact or a subjective judgment. ### Identify knowledge boundaries The agent must state its training cutoff and any data it cannot access to avoid guessing. ### Cross-reference for contradictions It checks the entire response to ensure paragraph 2 doesn't contradict paragraph 6. ## Use Cases ### Verifying Medical Research A researcher asks about a new drug. The agent uses validate_hallucination_grounding to ensure it doesn't just say "studies show" but provides the specific journal and study size. ### Financial Market Analysis An analyst checks market cap trends. The agent is forced to admit its knowledge cutoff or provide a specific source for 2024 data. ### Technical Documentation A developer asks about API limits. The agent identifies that it cannot verify real-time limits and states its knowledge boundary. ### Academic Literature Review A student summarizes a field of study. The agent separates the subjective analysis of the authors from the objective data points. ## Benefits - Stop Source Missing errors by forcing specific authors and DOIs for every claim using validate_hallucination_grounding. - Eliminate Opinion as Fact issues by requiring explicit labels on subjective judgments in your research. - Prevent Confidence Uncalibrated issues by scoring claims based on actual evidence quality instead of vague feelings. - Avoid Knowledge Exceeded errors by making the agent state its training limits clearly before it starts guessing. - Catch Self-Contradicting logic by cross-referencing all claims in a single response to ensure internal consistency. ## How It Works The bottom line is you get verifiable evidence instead of plausible fiction. 1. Connect the Hallucination Detector Prover to your AI client through the Vinkius dashboard. 2. Prompt your agent to perform a research task or verify a specific set of data points. 3. Receive a structured response that includes citations, confidence scores, and truth labels. ## Frequently Asked Questions **How does Hallucination Detector Prover stop my AI from lying?** It forces the AI to provide specific citations, like authors and DOIs, for every claim. If it can't find a real source, it is required to admit that it doesn't know the answer rather than guessing. **Can I use Hallucination Detector Prover for medical research?** Yes, it is ideal for medical research because it forces the agent to cite specific journals and peer-reviewed studies, helping you avoid fabricated medical claims. **How does it handle opinions vs facts?** The Connector requires the agent to explicitly label every statement. If the AI provides a subjective judgment, it must be tagged as an opinion so you can distinguish it from verifiable data. **Does Hallucination Detector Prover check for internal contradictions?** Yes, it cross-references claims across the entire response. This ensures that a fact stated in the beginning of a summary doesn't get contradicted in a later paragraph. **Why should I use Hallucination Detector Prover instead of just asking for sources?** Standard prompts often result in the AI making up fake sources. This Connector uses a structured reflection tool to ensure the citations provided are actually verifiable and grounded in reality. **What happens if the AI doesn't know a fact?** Instead of hallucinating, the AI is forced to state its knowledge boundaries. It will tell you if the information is outside its training data or if it lacks the necessary access to provide a certain fact. **What counts as a verifiable source?** Author or organization, publication name, date, and DOI or URL. 'Studies show' is rejected. 'Smith et al., Nature 2024, doi:10.1038/...' is accepted. **How does confidence calibration work?** The engine requires per-claim confidence with evidence quality: '90% confident (3 peer-reviewed sources)' instead of 'definitely' or '100% certain'. **Can it detect self-contradictions?** Yes. It rejects circular self-validation like 'as I said' and demands explicit cross-referencing by paragraph and claim number.