Social media platforms have become key players in political discourse. Twitter (now ‘X’), for example, is used by many German politicians to communicate their views and interact with others. Due to its nature, however, social networks suffer from a number of issues such as offensive content, toxic language and hate speech. This has attracted a lot of research interest but in the context of political discourse there is a noticeable gap with no such study specifically looking at German politicians in a systematic way. We aim to help addressing this gap. We first create an annotated dataset of 1,197 Twitter posts mentioning German politicians. This is the basis to explore a number of approaches to detect hate speech and offensive language (HOF) and identify an ensemble of transformer models that achieves an F1-Macros score of 0.94. This model is then used to automatically classify two much larger, longitudinal datasets: one with 520,000 tweets posted by MPs, and the other with 2,200,000 tweets which comprise posts from the public mentioning politicians. We obtain interesting insights in regards to the distribution of hate and offensive content when looking at different independent variables.