Purpose Multidisciplinary team (MDT) meetings are essential in Hepatobiliary (HPB) oncology with increasingly complex decision making. Accurate and efficient documentation of MDT discussions is important for clear communication and actionable outcomes. Large language model (LLM) based medical scribes such as Heidi Health, may improve communication and improve efficiency from the MDT by automating transcription and summarisation. Methodology This single-centre observational comparative study assessed the performance of Heidi in generating HPB MDT summaries compared with traditional documentation. Fifty consecutive cases were recorded, 35 were analysed following exclusion of poor quality audio. Each anonymised case was processed in two phases: (1) AI output aligned with the institutional MDT template and (2) AI self structured summary. Three HPB surgeons independently scored accuracy, clarity, conciseness, and actionability (0-10 scale). Results AI summaries demonstrated comparable quality to traditional MDT documentation (median score 7.0, p = 0.678). Accuracy of AI scribed outcomes were comparable to traditional MDT transcription in phase 1 (6.76 +/- 2.73) and 2 (6.87 +/- 2.58). Hallucination frequency was low (mean 0.25, median 0). Clarity was highly rated (phase 1 = 7.68, phase 2 = 7.83). Documentation speed improved markedly (122.5s vs 60s, p<0.0001). Conclusion Heidi produced accurate, clear and actionable MDT summaries with half the time required for documentation. Human verification remains essential though LLM assisted documentation represents a feasible and safe adjunct to traditional MDT documentation and communications.