Microsatellite instability-high (MSI-H) is a critical biomarker for immunotherapy, yet primary resistance remains a significant challenge. Current MSI-H detection methods evaluate the proportion of MSI-H loci, termed molecular MSI-H score, which can be affected by intratumoral heterogeneity (ITH). To address this limitation, we propose evaluating MSI-H at the cellular level to improve the prediction of immunotherapy outcomes. Using bulk tissue (TCGA-CRC) and cell line (CCLE-CRC) datasets, we identified genes highly expressed in MSI-H and MSS samples. These signatures were applied to a single-cell RNA sequencing (scCRC) dataset for enrichment analysis, enabling classification of tumor cells into MSI-H, MSS, and microsatellite dual (MSD) clusters using a Gaussian finite mixture model. Validation showed that MSI-H and MSS enrichment scores were higher in mismatch repair-deficient (MMRd) and mismatch repair-proficient (MMRp) patients, respectively. Functional enrichment analysis revealed that MSI-H cells were associated with pathways such as carboxylic acid catabolism, inflammatory responses, and IL-6/JAK2/STAT3 signaling. We developed a cellular MSI-H signature using genes specifically expressed in the MSI-H cell cluster and transformed the scCRC dataset into a cell-type-specific pseudobulk expression matrix. Using this matrix as a reference, we performed reference-based deconvolution on TCGA-CRC data. We defined the deconvolution score of MSI-H cell as cellular MSI-H score. This score strongly correlated with the molecular MSI-H score (R = 0.55, P < 0.001) and showed modest correlations with macrophage (MoMac, R = 0.14) and CD8+ T-cell (R = 0.11). To investigate its potential for clinical application, we applied the cellular MSI-H signature to the BJ-cohort, comprising 97 immunotherapy-treated gastrointestinal patients sequenced with a 395-gene panel. The cellular MSI-H score was significantly higher in responders (P = 0.002), positively correlated with tumor reduction percentage (R = 0.29, P = 0.006), and associated with improved progression-free survival (PFS) (HR: 0.00, 95% CI: 0.00-0.31, P = 0.021). In summary, the cellular MSI-H score reflects the MSI-H cell level within a tumor and demonstrates superior accuracy compared to molecular MSI-H status in predicting immunotherapy response and PFS. This underscores its potential as a more robust biomarker for guiding immunotherapy decisions.