<p>Magnetotelluric (MT) inversion is an inherently ill-posed problem, requiring the application of regularization methods to ensure stable and reliable solutions. The L2-norm regularization is widely used in electromagnetic inversion, typically yielding smooth models. However, these solutions often fail to reconstruct complex subsurface features accurately. To enhance the resolution of geoelectrical boundaries, non-L2 norm regularization approaches, regularization terms such as the minimum support (MS) and minimum gradient support (MGS) are considered. However, real-world geology often combines smooth and blocky features, which neither L2 nor non-L2 norm regularization alone can fully capture. To address this challenge, we introduce a mixed Lp-norm regularization, which combines the strengths of both L2 and L1 norms for two-dimensional (2D) MT inversion. Additionally, we propose an adaptive weighting scheme for L1 and L2 norm regularization, optimizing their contributions based on structural characteristics. To evaluate the effectiveness of our approach, we conducted inversion tests using two synthetic models, comparing the mixed Lp-norm regularization with conventional L1- and L2-norm approaches. The performance of our adaptive weighting scheme was also assessed against other weighting strategies. The results demonstrate the advantages of the proposed algorithm. Finally, we applied our method to the COPROD2 dataset to further validate its effectiveness in real-world scenarios.</p>