Web servers are the backbone of the modern Internet, handling user requests, managing data, and hosting applications, making them frequent targets for cyberattacks. In this context, HTTP logs are valuable source of information for analyzing and identifying potential threats. They have been used in Host-Based Intrusion Detection Systems (HIDS) to identify cyberattacks, but evolving threats demand ongoing research and new solutions. In this paper, a Multi-agent host-based Intrusion Detection and Prevention System (MIDPS) is proposed. It analyzes HTTP logs to detect various categories of cyberattack, such as file disclosure, SQL injection, and (D)DoS attacks. The proposed strategy is adaptive and independent of the analysed web application, as MIDPS automatically crawls the input application for new contents, recomputes and updates its threat detection heuristics. This is due to its adoption of a multi-agent approach, enhanced by the integration of Machine Learning. Moreover, when deployed on servers hosting web applications, MIDPS neutralised 100% of incoming File Disclosure, SQL injection, and (D)DoS attacks.