I wonder if integrating Model Context Protocol (MCP) could improve the accuracy of the clinical notes Heidi generates. If Heidi could securely access up-to-date guidelines and our local formularies, for example, the notes would be much more precise and reliable. Of course, I'm aware that allowing an AI to connect to external databases raises security concerns. Perhaps limiting MCP connections to approved secure servers, internal systems, or carefully vetted resources would be a way to manage this. Alternatively, would it be feasible to explore MCP calls that strictly do not involve patient-identifiable data?