Today, multi-agent technologies are widely used for realizing prevalent ubiquitous computing applications, in which service discovery is a critical task for finding a particular service instance. While JADE is a popular system for multi-agent applications, its directory facilitator (DF) used for service discovery employs a sequential search approach, which shows degraded performance when the number of registered services becomes large. This paper proposes a new DF scheme employing the category-based classification and search approach. It greatly reduces the search space and allows accurate matchmaking. The DF implemented with the proposed approach and JADE-DF are compared in terms of query response time and memory space requirement. It demonstrates that the proposed DF allows faster query processing than JADE-DF and requires smaller memory space, especially for a large number of services and queries of multiple parameters.