We propose a novel MOE-based architecture that combines the detector and the visual foundation model for cross-domain few-shot object detection.