Sign language recognition is a crucial element in the development of accessible communication technologies, particularly for addressing communication barriers between hearing and non-hearing individuals. This study presents a unified approach for recognizing the complete Mexican Sign Language (LSM) alphabet, which includes 21 static and 6 dynamic signs. The methodology involves extracting keypoints from video frames captured from 20 subjects, considering different frame lengths. Machine learning models were trained and evaluated on both static and dynamic signs, assessing performance for each letter individually and by groups. The model using the starting and middle frames achieved the best performance (efficacy and efficiency), with an F1-score of 0.97. The unified framework aims to support continuous and real-time recognition, enhancing tools for inclusion and reducing communication gaps with the Deaf community.