A geometric derivation of numerical integrators for optimal control problems is proposed. It is based in the classical technique of generating functions adapted to the special features of optimal control problems.This work has been supported by grant BFM2001-2272 (Ministry of Science and Technology, Spain). A. Santamaría Merino wishes to thank the Programa de Formación de Investigadores of the Departamento de Educación, Universidades e Investigación of the Basque Government (Spain) for financial support.Peer reviewed