Heterogeneity in open fMRI datasets, caused by variations in scanning protocols, confounders, and population diversity, hinders representation learning and classification performance. To address these limitations, we propose a novel multi-site adversarial learning network (MSalNET) for fMRI-based mental disorder detection. Firstly, a representation learning module is introduced with a node information assembly (NIA) mechanism to extract features from functional connectivity (FC). This mechanism aggregates edge information from both horizontal and vertical directions, effectively assembling node information. Secondly, to generalize the feature across sites, we proposed a site-level feature extraction module that can learn from individual FC. Lastly, an adversarial learning network, is proposed to balance the trade-off between individual classification and site regression tasks. The proposed method was evaluated on Autism Brain Imaging Data Exchange (ABIDE). The results indicate that the proposed method achieves an accuracy of 75.56% with reducing variability from a data-driven perspective. The most discriminative brain regions revealed by NIA are consistent with statistical findings, uncovering the black box of deep learning to a certain extent. MSalNET offers a novel perspective on the detection of multi-site fMRI mental disorders and it considers the interpretability of the model, which is a crucial aspect in deep learning.