This paper explores the use of information agents for partially automating some of the activities of information brokering in e-commerce. In this paper, the issue of matching buyers’ and sellers’ profiles is viewed as a kind of connection problem. A connection algorithm that matches advertisement of buyers and requests from sellers based on pre-specified multiple criteria was devised and implemented. The connection algorithm mainly consists of: selection of requests and advertisement, and evaluation and filtering of connected advertisement and requests. Ideas of the connection approach are realized in a testbed on securities trading consisting of a society of information agents (a broker agent, a recommendation agent and a record agent) and trading agents (buyer and seller agents) that communicates via a blackboard database. In addition, an information brokering protocol was devised and implemented to structure the interactions and information exchange among agents in the testbed. The protocol does not assume that connections of advertisements and requests can always be made. When the broker agent fails to match requests and advertisements, the recommendation agent proactively pushes e-mail recommendations to potential trading agents by tracing previous transaction profiles. A series of experiments that were carried out shown favorable results in executing the protocol.