The development of refined tropical cyclone (TC) forecasting operations relies on high-resolution regional TC numerical forecasting. However, when faced with both the characteristic of expensive computational consumption of high-resolution models and the characteristic of the wide range of TC forecasting responsibility areas, it is difficult for the existing numerical forecasting techniques to meet the high timeliness requirements in operational high-resolution TC forecasting. High-resolution numerical models face challenges in balancing computational demands, forecast timeliness, and prediction accuracy for TC forecasting operations. To save computational resources and meet high timeliness requirements while ensuring TC forecast accuracy, this study proposes an adaptive setting-up approach of a limited area for the short-term forecast of TC by combining an adaptive objective calculation method for the TC simulation domain with multi-TC coexistence processing technology. This approach dynamically adjusts the position and size of the high-resolution TC model simulation domain based on the forecast timeliness and the actual situation of the TC. By automatically providing a scientifically reasonable simulation domain, the computer resource consumption is reduced without losing forecasting skills. Applying this approach to the high-resolution CMA-MESO model, a High-Resolution Typhoon Numerical Prediction System (HRTYM) is established. The numerical experiments and comparative analysis are carried out using Typhoon Lekima (2019) and 16 major typhoon events in 2020 and 2021. The experimental results of Typhoon Lekima show that, compared with the 9 km-resolution operational model (CMA-TYM), the 3 km-resolution HRTYM requires fewer computational resources with model integration time reduced by 11.2%-17.5% and storage space reduced by 58.6%-66.7%, the TC track forecast error decreases by 20.7%-61.0%, and the TS scores for 24-hour rainstorm and heavy rainstorm predictions improve significantly. The results of batch experiments in 2020 show that after 24-48 hours of mode integration, the track error of the 3 km-resolution HRTYM decreases by 3.44-34.91 km compared to the 9 km-resolution CMA-TYM, and the results of batch experiments in 2021 show that after 27-48 hours of mode integration, the track error of the 3 km-resolution HRTYM decreases by 0.6-22.35 km compared to the 9 km-resolution CMA-TYM. The results of batch experiments in 2020 and 2021 demonstrate that HRTYM exhibits superior track prediction skill over CMA-TYM beyond 27 hours. The high-resolution TC numerical prediction system applying the adaptive setting-up approach of a limited area for the short-term forecast of TC proposed in this study effectively reduces the cost of computing resources and computer storage space resources and ensures the skill of TC numerical prediction at the same time.