Commercial IoT search engines, such as Shodan, are costly and have several limitations, primarily due to their limited data analysis capabilities and restrictions on the use of their services. This work explores the feasibility and utility of an alternative approach. A prototype IoT search engine was developed capable of acquiring, indexing, retrieving, classifying, and analyzing textual banners associated with publicly accessible devices on the Internet. The system predominantly integrates open-source software but still relies on Shodan for banner collection on the Internet. It allows for detailed analysis and large-scale experimentation on devices of interest without the constraints of commercial systems. For the project’s goals, the problem of identifying and characterizing network-accessible devices in the energy sector was studied. This problem presents a series of conceptual and practical difficulties and is not sufficiently addressed in the literature. The work specifically focused on the problem of classifying surveyed devices using machine learning methods. Shodan’s labeling process, in fact, has low coverage and various anomalies. Through careful engineering of the training set (also conducted by automatically extracting relevant technical information from banners), a Random Forest classifier was developed that achieved high performance in tests on ICS (Industrial Control Systems) devices, including those in the energy sector. The prototype represents an important tool for future security applications in the energy sector. The results achieved in recognizing ICS-type banners and automatically extracting technical information from them (albeit still partially) make it a useful support tool for assessing the vulnerability of such devices to cyberattacks.