Recently multi-agent system (MAS) is regarded as a powerful paradigm to develop and simulate complex system. Because supply chain (SC) is a complex network, it is hard to improve supply chain management (SCM) efficiency and automate SC operating process. Researchers have applied MAS in SCM viewing enterprises in SC as different functional agents collaborating to achieve common goal. In this paper a MAS-based supply chain management framework (MSF) is developed on which agents can collaborate to conduct business activities of enterprises and customers. Six kinds of functional agents are designed, namely, management agent, order receiving agent, order decomposing agent, resource agent, manufacturing-controlling agent and transportation agent. Through the agents collaboration the MSF can support SC dynamic forming and operating process. For partner selection the method to transmit customers order into demand matrix, and transmit enterprises resource capacity into resource matrix is described in detail. The MSF users decide which algorithm to be applied in matching the two matrixes, and algorithms are stored in model base. Finally the generic agent architecture is put forward.