Electroencephalography-based brain-computer interface (EEG-BCI) systems have been developed to enable individuals with physical disabilities to control external devices using their thoughts. Motor imagery (MI) is one of the most common paradigms used in EEG-BCI, where users are instructed to imagine performing a specific motor task, such as moving their left or right hand, without physically performing it. Using fewer electrodes in EEG recordings has several advantages for real-world applications and for widespread use. It reduces the complexity and cost of the recording setup. Moreover it improves user comfort, allowing for more flexibility and ease of use in various settings, such as home-based or mobile applications. However, using fewer electrodes can also result in lower signal quality and reduced spatial resolution, leading to lower classification accuracy and decreased robustness to artifacts and noise. Another critical issue in BCI performance is the amount of training data available, leading to increasing problems of overfitting. The Augmented Covariance Method (ACM) is a technique that has been able to improve the classification accuracy of EEG-BCI. During this study we focus on applying this methodology in a setting with low number of electrodes (specifically with three electrodes C3, C4, and Cz) and low number of training sample. This study evaluates the effectiveness of ACM on several datasets from the MOABB package, focusing on different classification tasks and evaluation procedures. Even with a small number of electrodes and limited training samples, the Augmented Covariance Method (ACM) is proven to be effective in BCI classification, demonstrating its potential to enhance the performance of BCIs in real-world applications. For example in the right hand vs left hand task classification with dataset BNCI2014001 using the within-session evaluation we obtain an Area Under ROC Curve (AUC) of 0.86, compared to results of 0.95 using all electrodes available. These findings suggest that ACM can be a valuable tool for improving the usability and effectiveness of BCIs, particularly in scenarios where only a limited number of electrodes are available.