The Goal Decomposition Tree model has been introduced in 2005 by Mermet et al. [9] to specify and verify the behaviour of an agent evolving in a dynamic environment. This model presents many interesting characteristics such as its compositional aspect and the definition of proven proof schemas making the proof mechanism reliable. Being interested in specifying and verifying multiagent systems, we have decided to extend the GDT model for specifying Multiagent systems. The object of this article is to present this extension. So, after a brief description of the initial GDT model, we show how we extend it by introducing the specification of the whole MAS. We also introduce the notions of agent type and agent, and we show how external goals allow to specify collaborative agents and to prove the correctness of their collaboration. These notions are illustrated on a toy example of the litterature. Categories and Subject Descriptors D.2.4 [Software engineering]: Software/Program Verificat...