To meet the specific requirements of Multi-Agent Systemtask allocation coordination, this paper puts forward an extended contract net model based on mental state constraints. The suggested mental state includes mental factors such as trust degree, loyal degree, active degree, risk tolerance degree, busy degree and urgency degree. By setting rational mental state parameters, it narrows down bidder agents to a rational and limited number, reducing significantly system communication cost. The mental state is applied from both the perspective of the manager agent and that of the bidder agents and it is more rational by employing different adjustment coefficients according to the property of the task. The system control employs a mixed structure of centralized and distributed control to reduce complexity. The overall task allocation coordination enjoys finer granularity by using ability and resource required by the task, instead of the whole task, as the basis of operation. Finally, the example of emergency relief support task allocation is descripted to validate the effectiveness of the suggested model.