This paper studies the principles and algorithms of non-structural P2P auto search based on peer interests. The interests of nodes represented by a term-weight vector of shared documents. With pull-only communication, peers in the network cannot push messages to other peers. The messages will be transferred only by requests. Nodes in the system can analyze the user's shared documents, and automatically know the type of document of interest to users. Nodes with similar interests can automatically form an overlay network. Node can periodically and automatically access other nodes in the network coverage of the new documents. So convenient for users to find and view the new documents. Simulation result show the system is scalable.