Abstract: The application of large language models (LLMs) in the field of mental health has become a key direction in the interdisciplinary research between artificial intelligence and clinical psychology. This review systematically summarizes recent developments from three perspectives: theoretical foundations, clinical applications, and technological advancements. At the theoretical level, it analyzes the intrinsic characteristics of LLMs and outlines the rationale for their suitability in psychological symptom diagnosis and therapeutic intervention. At the application level, it reviews practical cases and outcomes of LLMs in mental disorder diagnosis, psychological state assessment, virtual therapy, and clinical decision support. At the technological level, it synthesizes key progress in data construction, capability enhancement, and evaluation methods tailored to mental health contexts. Finally, the review identifies current challenges, including the gap between diagnostic modeling and clinical practice, the limited depth of therapeutic simulation, the scarcity of high-quality data, and the lack of clinical validation for technological approaches. It concludes by discussing future directions for integrating LLMs into clinical applications and advancing technical research in the mental health domain.