Semaphore is a communication method used to send and receive messages by utilizing flags, paddles, sticks, bare hands, or gloves. Semaphore signaling is commonly used in scouting activities (Pramuka). However, novice scouts often face difficulties in accurately recognizing semaphore gestures. Although guidebooks are available, errors may still occur, especially when learning independently. To address this issue, a gesture detection system using the Multi-Layer Perceptron (MLP) method and MediaPipe framework is proposed, which aims to evaluate the accuracy of MLP and MediaPipe in detecting semaphore poses. To assess the performance of the MLP model, a Confusion Matrix is used to analyze the number of correct and incorrect predictions and generate metrics such as accuracy, precision, recall, and F1-score. The evaluation results show that the system successfully recognized 122 poses correctly, achieving an accuracy of 93.84%, precision of 93.84%, recall of 100%, and an F1-score of 96.80%. Based on these results, the system is considered sufficiently accurate and can be used as an interactive learning tool to recognize semaphore codes, particularly in scouting activities.