Abstract: The uncertainty of measured cross sections is relatively large due to the higher energy of neutrons in fast neutron spectrum reactors and the smaller cross sections compared to the low-energy range. The uncertainty of these cross-sections affects the uncertainty of reactor physics calculations results. Previous research results show that nuclear data uncertainty is a major source of uncertainty in reactor physics calculations. However, the analysis of uncertainty requires software for processing nuclear cross section covariance data which are from the neutron Evaluated Nuclear Data File (ENDF), which heavily relies on relevant foreign computational software, such as NJOY. Currently, there are few reports on domestic software for processing nuclear data covariance data, and there is limited research on its methods and theories. Therefore, this study delves into the theory of multi-group covariance ENDF data processing method and develops the multi-group covariance processing module CovarXS within the independently developed advanced nuclear cross section (XS) processing program AXSP. A 33-group covariance database was generated based on the ENDF/B-Ⅶ.1 evaluated nuclear data file. To verify the correctness of the developed CovarXS module, relative covariance matrices for ²³⁸U(n,inel), ²³⁹Pu(n,fiss), and ⁵⁶Fe(n,elas) were generated using the module based on the ENDF/B-Ⅶ.1 evaluated nuclear database. By comparing the results with the ERRORR module in NJOY2016, it is found that the maximum relative error of the computed covariance data between the two programs is within 1.0%. The relative errors for ²³⁸U(n,inel) and ²³⁹Pu(n,fiss) are less than 0.001%. The relative error for ⁵⁶Fe(n,elas) is slightly larger, but still do not exceed 1%. Based on these results, the 33-group covariance matrices were used to analyze the BN-600 benchmark problem of a MOX-fueled sodium-cooled fast reactor. The nuclides and types of cross-sections that contributed the most uncertainty to the k_eff were identified. The calculation results show that the relative errors of the total k_eff uncertainty contributions from all nuclides obtained using the CovarXS module and the ERRORR module are less than 0.11%. The nuclide with the largest contribution is ²³⁸U, followed by ²³⁹Pu, ⁵⁶Fe, ²³Na, and others. For the ²³⁸U nuclide, the reaction channel contributing the most to k_eff is (n,inel), followed by the (n,g) reaction channel of ²³⁹Pu. The total relative uncertainty of k_eff caused by the main nuclides is 0.893%. This indicates that the precision of the multi-group covariance database generated by the CovarXS module is comparable to that generated by the ERRORR module. The relative covariance data generated using the CovarXS module can provide reliable data for subsequent uncertainty analysis.