Detecting brain tumors is crucial in modern healthcare for timely diagnosis and treatment. This study introduces an IoT-based convolutional neural network (CNN) framework incorporating separable convolution layers to enhance the accuracy and efficiency of brain tumor detection in MRI images. Designed for IoT-enabled environments, the model combines standard and separable convolution layers to balance precision and computational efficiency, addressing real-time processing and edge computing challenges. Using depthwise separable convolutions and efficient pooling layers, the framework achieves detailed feature extraction while reducing complexity. Trained on the Br35H brain tumor dataset with data augmentation, the model achieves a remarkable accuracy of 99.17%, outperforming traditional CNNs in precision and computational load.