The minimally invasive surgery (MIS) has gained widespread popularity in both robot-assisted and conventional surgical procedures because of its relative ease of use and minimal invasiveness. However, limited visibility through small incisions poses challenges for surgeons, and the use of robotic systems may risk organ damage if surgical tools are not accurately identified in real time. Consequently, reliable and rapid detection of surgical tools is crucial for ensuring safety and efficacy in MIS and robotic surgeries. Current deep learning-based detection techniques often face a trade-off between detection speed and accuracy, compounded by their complex model architectures and substantial hardware requirements, which hinder practical deployment. This paper presents a lightweight surgical tool detection model integrating YOLOv11 with MobileNetV3, aimed at enhancing detection accuracy while addressing challenges such as low precision, large model size, and deployment difficulties. The proposed approach incorporates the MobileNetV3 architecture into the YOLOv11 framework to improve computational efficiency. Additionally, we introduce the Shape-IoU loss function to enhance the model’s detection and bounding box regression accuracy. Experiments conducted on the open-source m2cai16-tool-locations laparoscopic dataset demonstrate that our method surpasses existing approaches, achieving 95.1% mAP0.5 and offering a 28% increase in detection speed.