A multi-agent system (MAS) is an automated system which consists of multiple agents. MAS can be used to solve the problems that are difficult to solve by human. Agents are connected with each other; they have access to other agent via connection. The problems and functions are not stored automatically in agent, agents uses their connections for storage. High storage in connections may leads to occur memory insufficient. This memory insufficient also creates the changeability of apparent input. This memory insufficient has allowed us to define the model of MAS by using markov property. Here we use markov chain model to define strength of MAS. The number of states varies according to the capability of original agents. We then explore the strength of evolving agent growth through simulation and compare the results with non-derive MAS. We can implement classified agents in data extraction from cloud data server.