The goal of this project is to model a MAS -- multi-agent web system -- which integrates other expert systems, represented in this application as agents, through a web server. To this end, we developed a prototype of a MAS called CLIDENP -- Virtual Clinic for the Diagnosis and Education of Pediatric Nutrition. CLIDENP allows the diagnosis of nutritional states and suggests diets for children up to two years of age. The results demonstrate that the application of a MAS is capable of facilitating the exchange of messages and services between the original expert systems, by means of a dynamic and autonomous communication language similar to Knowledge Query and Manipulation Language -- KQML. In this way, the proposed model permits a set of expert systems to work in a distributed fashion and solve problems in a complex domain.