In this paper, a planning-based approach to service composition in medical domain is described. A service composition problem is presented as the syntactic and semantic matching problem, where services are chained together so that one service output can be used in another service input, and in the end provide user with requested goal outputs. Focus is put on the planning aspects of the problem and searching the planning space with the use of modified GraphPlan algorithm. Modifications focus on implementation of a detailed model of service inputs and outputs. Finally, performance study is presented along with commentary on application of the algorithm in the medical domain.