Getting the Source Right, Not Just the Fact: Source-Aware Verification for MCP Agents

AI agents that use multiple tools can create a subtle but critical error where a stated fact is true but attributed to the wrong source. A new verification system called ProvenanceGuard was developed to solve this by ensuring not only that a claim is factually supported, but that it is supported by the specific source the AI cites. The system works by breaking down an AI's answer into individual claims and then meticulously checking each one against the original tool outputs without ever pooling the evidence. In tests on a medical agent, ProvenanceGuard proved highly effective at catching these attribution errors, correctly blocking nearly every single claim that was improperly sourced.