In recent years, with the improvement of computer performance and numerical calculation methods, reactor calculation tools have also developed in the direction of high fidelity: high-precision computing has become a hot topic. Therefore, considering the maximum error of the optimal estimation, this paper combines nuclear-thermal coupling with uncertainty analysis for a more accurate study. In order to reduce the time-consuming and human error of the coupling process, this paper develops an MCNP-Fluent coupling interface script based on Python language, which automatically completes the data exchange between programs, and greatly improves the efficiency of calculation and the accuracy of the results. Based on this interface, the nuclear thermal coupling calculation is carried out for the maximum power fuel assembly of the Minjiang Reactor (MJTR): the total power of the fuel assembly is calculated by MCNP, and the heat source is loaded into the thermal calculation model in the form of UDF. The results show that there is little difference between the peak power before and after coupling, and the peak point is offset by about 3 cm towards the inlet. The maximum temperature of each layer of the fuel after coupling is slightly lower than before coupling, and the maximum temperature is within the safe limit. Taking MJTR fuel assembly kinf as the target response, the sensitivity coefficients of each cross-section of 1H, 16O, 235U and 238U nuclides to kinf were calculated. Using the high-efficiency sampling method, the ACE format kernel data cross-section was made by the NJOY program for the MCNP program to calculate, and the uncertainty level of the kinf after the perturbation of the 1H capture cross-section was quantified, and the uncertainty level of the power response of the module caused by the 235U fission cross-section was further analyzed.