Server-centric data center architecture has been proposed to provide high throughput, scalable construction and error tolerance with commodity servers and switches for cloud data centers. To fully utilize those advantages of servercentric data center, an effective routing algorithm to find high quality multiple paths in Server-centric network is needed. However, current routing algorithms cannot achieve this completely: 1) the state-of-art routing algorithms in server-centric data center just consider hop count when selecting paths; 2) traditional multi-constraint QoS routing algorithms only find one feasible path and are usually switch-oriented; 3) present multi-path algorithms cannot guarantee the performance of the founded paths. In this paper, we propose a multi-constrained routing algorithm for servercentric data centers, named Server-Centric Multi-Constrained Routing Algorithm (SCRAT). This algorithm exploits the topology features of the Server-Centric data center to decrease the algorithm complexity and returns optimal and feasible paths simultaneously. In simulations, SCRAT has a very high probability (more than 96%) to find the exact optimal path, and the cost of the optimal path found in SCRAT is about 10% less compared with path found in previous TS MCOP. Compared with previous MPTCP, SCRAT reduces the path delay by 18% less and increase the bandwidth by 20%.