Existing incomplete multi-view clustering algorithms face several limitations, including the difficulty in preserving the original structure during missing instances recovery, and the inability to accurately capture local structures and high-order information in multi-view data. To address these issues, a multi-source information reconstruction and self-representation (MSRS) for incomplete multi-view clustering algorithm is proposed. The MSRS first reconstructs incomplete views that reflect the original data structure based on multi-source information. Then, based on the reconstructed views, the MSRS employs an effective regularization method that combines sparse constraints with local structure preservation, and introduces a weighted tensor ${\rm Schatten}\text{-}p $ norm to dynamically control the contributions of different singular values, thereby learning high-quality self-representation matrices for each view. Experiments comparing the MSRS with nine advanced baseline algorithms on various incomplete datasets demonstrate that, the proposed algorithm significantly outperforms the baselines in most cases.