Objective To enhance teaching in postoperative cancer rehabilitation, this study developed an integrative Chinese-Western medicine postoperative oncology rehabilitation system, termed the medical oncology generative pre-trained transformer (MedOncoGPT). By introducing MedOncoGPT as an intelligent assistant, an integrated teaching model combining Chinese and Western medicine was established. The study evaluated its impact on students’ integrative clinical reasoning and practical abilities, providing support for instructional reform in related courses.Methods Using teaching resources as the knowledge base, MedOncoGPT was built upon the open-source ChatGLM model and incorporated Low-Rank Adaptation (LoRA) fine-tuning and retrieval-augmented generation (RAG) techniques to address postoperative integrative oncology scenarios. The system was applied in courses and clinical clerkships related to integrative oncology. In alignment with course objectives, a five-stage instructional process—pre-class preparation, in-class inquiry, simulated multidisciplinary consultation, clinical reinforcement, and teaching reflection—was designed to guide students in completing syndrome differentiation, comprehensive assessment, and follow-up planning within real or simulated case contexts. Comparative analyses of student engagement, syndrome differentiation thinking, evidence-based awareness, and interdisciplinary integration skills before and after the teaching reform were conducted using questionnaires, course assessments, classroom observations, and semi-structured interviews.Results Following the implementation of MedOncoGPT, students demonstrated improved performance in case analysis, prescription formulation, and integrative Chinese-Western medical evaluation compared with those receiving traditional instruction. Classroom participation and the relevance of student inquiries also increased. Self-assessment results indicated high levels of satisfaction with respect to clarity of integrative clinical reasoning, ability to retrieve and apply guideline-based evidence, and awareness of appropriate use of intelligent tools in clinical decision-making. More than 92% of students reported that the system facilitated understanding of abstract theoretical concepts presented in textbooks. Instructors noted that the system helped reduce lesson preparation time, enriched typical case materials and discussion scenarios, and promoted the translation of research findings into classroom teaching. Pilot data showed that, with MedOncoGPT assistance, the mean time for initial syndrome differentiation decreased from 18.4 min to 12.1 min, and the agreement rate increased from 68.3% to 82.5%. In the teaching pilot, the experimental group achieved a higher mean score on the final case analysis assessment than the control group (82.6 vs. 74.3).Conclusion The integration of MedOncoGPT into teaching on postoperative integrative cancer rehabilitation enabled the establishment of a stable instructional process within existing curricula and enhanced students’ integrative clinical reasoning and evidence-based practice capabilities. The approach demonstrates positive potential for advancing the integration of research, clinical practice, and education and represents a valuable exploratory strategy for instructional reform in courses on integrative Chinese-Western medicine.