Recently, small-signal stability of the modular multilevel converter (MMC) based high-voltage direct current (HVDC) transmission for wind farm integration has attracted great attentions. The impedance-based frequency-domain method is an effective tool for such studies, in which the accuracy of the impedance model is of significant importance. Currently, the decoupled single-input single-output (SISO) sequence impedance of the MMC-HVDC system with wind farm is commonly used due to the simplicity in stability analysis. However, since both the MMC and wind farm exhibit frequency coupling behaviors, the decoupled SISO sequence impedance model may lead to inaccurate stability conclusion under certain conditions. In order to improve the model accuracy, based on the harmonic state-space (HSS) modeling approach, this paper proposes a generalized multi-input multi-output (MIMO) sequence impedance model of the MMC-wind farm system. The proposed MIMO impedance model can effectively capture the frequency couplings in the interconnected system, based on which the coupling mechanism of multi-frequency components within the MMC is then revealed. To facilitate the interconnection analysis, the model truncation criteria is proposed and validated by comparing the accuracy of the MIMO model with different truncated dimensions. In addition, this paper considers the internal impedance network of the wind farm when developing the aggregated wind farm impedance model for interaction stability analysis. Finally, the proposed MIMO sequence impedance model is applied to accurately predict the instability of the MMC-HVDC connected wind farm system.