Tumor-derived small extracellular vesicles (tEVs) as potential biomarkers possess abundant surface proteins closely related to parent cells, which are crucial for noninvasive cancer diagnosis. However, tEVs exhibit phenotype heterogeneity and low abundance, posing a significant challenge for multiplex detection with a high sensitivity. Herein, we developed a DNA gate-based exponential amplification CRISPR-Cas (DGEAC) system for accurate and ultrasensitive detection of tEVs, which can greatly improve the accuracy of breast cancer (BC) diagnosis. Based on the coexpression of CD63 and vascular endothelial growth factor (VEGF) on BC-derived tEVs, we developed a dual-aptamer-based AND gate fluorescent probe by proximity hybridization. By integrating the target recognition and trans-cleavage activity of Cas12a, an autocatalysis-driven exponential amplification circuit was developed for ultrasensitive detection of CD63 and VEGF proteins on tEVs, which could avoid false negative signals from single protein or other interfering proteins. We achieved highly sensitive detection of tEVs over a linear range from 1.75 × 103 to 3.5 × 108 particles/mL with a detection limit as low as 1.02 × 103 particles/mL. Furthermore, the DGEAC system can distinguish tEVs from tEVs derived from different BC cell lines, including MDA-MB-231, MCF-7, SKBR3, and MCF-10A. Compared to linear amplification (AUC 90.0%), the DGEAC system effectively differentiates BC in different stages (AUC 98.3%).