Tumor MT is a novel method to identify tumor mesenchymal transition status. In our work, we developed a MTCG list containing 35 genes and explored MT variations in 9,415 patients from 24 types of epithelial cancers in The Cancer Genome Atlas dataset. Cancer samples were robustly classified into four MT clusters (MTlow, MTmid, MThigh-Hippo and MThigh-AKT) with distinct molecular, biological and clinical features. The enhanced MT status in MThigh-Hippo and MThigh-AKT types were found to be driven by intrinsic genetic alterations or extrinsic microenvironment interplays, respectively. Importantly, interplay between tumor cells and macrophages in MThigh-AKT patients not only facilitates MT process, but also leads to the dysfunction of antitumor T cells, which should be blocked to overcome the common failure of ICB treatment in these patients. Based on this, we developed this Tumor MT website, an interactive web-based resource that makes uses of transcriptomic data to conduct tumor MT classification and provide reasonable therapeutic options.