In order to improve the performance of network management task, the mobile agent based complex network management task cooperation mechanism is analyzed. It can be found that the traditional agent collaboration models (such as the contract-net protocol) are not suit for the large scale complex task agent cooperation and the performance of an individual agent will not be guaranteed. The cooperative game theory is introduced to solve the above problems. In this distributed collaboration model, an individual agent is regarded as a self-interest agent and it has its own utility function to evaluate the performance of the collaboration model. A convex coalition game model is proposed and the Shapley value of cooperative game theory is proved to be the criteria of fair task allocation in the collaboration model. Based on the above theory model, a three-stage task negotiation algorithm is proposed.