The changes in gut microbiota have been implicated in colorectal cancer (CRC). The interplays between the host and gut microbiota remain largely unclear, and few studies have investigated these interplays using integrative multi-omics data. In this study, large-scale multi-comic datasets, including microbiome, metabolome, bulk transcriptomics and single cell RNA sequencing of CRC patients, were analyzed individually and integrated through advanced bioinformatics methods. We further examined the clinical relevance of these findings in the mice recolonized with microbiota from human. We found that CRC patients had distinct microbiota compositions compared to healthy controls. A machine-learning model was developed with 28 biomarkers for detection of CRC, which had high accuracy and clinical applicability. We identified multiple significant correlations between genera and well-characterized genes, suggesting the potential role of gut microbiota in tumor immunity. Further analysis showed that specific metabolites worked as profound communicators between these genera and tumor immunity. Integrating microbiota and metabolome perspectives, we cataloged gut taxonomic and metabolomic features that represented the key multi-omics signature of CRC. Furthermore, gut microbiota transplanted from CRC patients compromised the response of CRC to immunotherapy. These phenotypes were strongly associated with the alterations in gut microbiota, immune cell infiltration as well as multiple metabolic pathways. The comprehensive interplays across multi-comic data of CRC might explain how gut microbiota influenced tumor immunity. Hence, we proposed that modifying the CRC microbiota using healthy donors might serve as a promising strategy to improve response to immunotherapy.