We consider mechanism design problems with n agents when the mechanism designer cannot fully commit to an allocation function. With a single agent (n=1) optimal mechanisms can always be represented by direct mechanisms, under which each agent’s message set is the set of his possible types [Bester, H., Strausz, R., 2000. Contracting with imperfect commitment and the revelation principle: the ... [Show full abstract] single agent case. Free University of Berlin, mimeo]. We show that this result does not hold if n≥2. That is, in mechanism design problems with multiple agents the use of direct mechanisms may be suboptimal.