Anonymity likes company. Hence peer-to-peer networks, represented by Tarzan [14] and MorphMix [26], have become a significant paradigm for anonymous communications because they are scaleable, self-organizing and reliable. However, little work has been done on how to build a peer-to-peer anonymous overlay network. To address this problem, we propose the following: (1) Probability Name-Dropper based upon node local In-degree (PNDI), a peer discovery scheme using only local information instead of global to build a peer-to-peer mix overlay; (2) a topology optimization scheme for resolving mismatch between the logical topology and the physical topology; (3) four local path selection strategies partially aiming at evaluating our peer discovery scheme. Simulation results show that this overlay presents scale-free and small-world properties, and our topology optimization scheme can effectively solve the mismatch problem and significantly reduce the traffic.