Objective: Advanced gastric cancer remains highly refractory to therapy, with limited immunotherapy efficacy due to tumor microenvironment heterogeneity. Primary cilia, microtubule-based organelles involved in tumor progression, remain insufficiently explored in gastric cancer. This study aimed to define primary cilia subtypes and establish prognostic signatures for personalized treatment strategies. Methods: Bulk transcriptomic data from over 1,500 gastric cancer samples were integrated to define distinct primary cilia subtypes. A primary ciliary phenotype-associated signature (PCS) was established using a multi-machine learning survival framework incorporating ten algorithms. The prognostic predictive value and immunotherapy response prediction capability of PCS were validated across multiple independent cohorts. Single-cell RNA sequencing analysis was performed to identify cellular populations associated with high-PCS phenotype. Causal weighted gene co-expression network analysis (WGCNA) was employed to identify driving factors, followed by functional validation through cell culture experiments and xenograft models. Results: Two distinct primary cilia subtypes were identified and validated across all cohorts, with C2 patients exhibiting significantly worse overall survival compared to C1 patients. PCS demonstrated robust predictive value for both prognosis and immunotherapy response, with superior accuracy compared to existing models across multiple validation cohorts. High-PCS patients showed reduced tumor purity, increased stromal cell infiltration, and poor response to immunotherapy. Single-cell analysis revealed that fibroblasts had the highest PCS scores and identified a novel secreted modular calcium-binding protein 2 ( Conclusions: PCS serves as a robust prognostic biomarker for gastric cancer patients. Additionally, targeting