In recent years, artificial intelligence–especially large-scale foundation models–has advanced rapidly in healthcare. Traditional Chinese Medicine (TCM) diagnosis relies on the four examinations (inspection, listening/smelling, inquiry, and palpation) and is highly dependent on clinician experience, leading to subjectivity and variability. The emergence of multimodal large language model (MLLM) offers new opportunities for making TCM diagnosis more objective and standardized. This review surveys the latest progress on applying MLLM and large language models to TCM-aided diagnosis. Focusing on three representative diagnostic dimensions–tongue, pulse, and facial diagnosis–we summarize how image analysis, sensing technologies, and natural-language processing enable data acquisition, preprocessing, and analytics for these modalities. We then synthesize current applications and performance of LLMs in TCM diagnostic settings, with particular attention to systems that integrate tongue/facial images, voice-based consultations, and textual medical records for multimodal, AI-assisted decision making. We further highlight advances and open challenges in applying LLMs to multi-visit, longitudinal follow-up scenarios, and discuss how such models may enhance the scientific rigor, standardization, and verifiability of TCM clinical practice. Finally, we outline future research directions–including cross-modal alignment, knowledge enhancement, personalized prediction, and large-scale multimodal pretraining–to inform subsequent research and applications.