Our previous study identified significant differences in tongue images between individuals with and without gastric cancer, but the underlying biological mechanisms remain unclear. In this study, we employed pressure cycling technology and data-independent acquisition (PCT-DIA) mass spectrometry to acquire tongue-coating metaproteomics data from 180 gastric cancer patients and 185 controls across five centers in China. We characterized 1,432 human proteins and 13,780 microbial proteins in these tongue-coating samples and demonstrated the temporal stability of protein expression levels using a time-series cohort. We also compared the functional and taxonomic characteristics of the tongue-coating proteomes between individuals with and without gastric cancer, identifying notable alterations in gastric cancer patients. Notably, keratins KRT2 and KRT9, and the microbial ABC transporter COG1136 were downregulated in gastric cancer patients, indicating reduced lingual mucosa defense. Importantly, we developed a machine learning model using 50 microbial proteins to identify high-risk individuals, achieving an AUC of 0.91 in the independent validation cohort. This study highlights the potential of tongue-coating proteins as indicators for non-invasive gastric cancer screening.