This paper suggests an ant-like agent service discovery mechanism. There are two types of agents cooperating to search target services: Search Agent and Guide Agent. Search Agent simulates the behavior of an ant that searches for services on the network. Guide Agent is responsible to manage a service route table that consists of pheromone and hop count, instructing Search Agent’s routing. Volatile pheromones make Search Agent sense the change of topology and service resource, and hop count makes them know the distance. Semantic similarity is also introduced in routing selection as a heuristic factor, which improves the recall. The life-span control policy makes query traffic controllable. With system size increasing, the query traffic would increase slightly and has an upper bound. The result of simulation shows that the suggested mechanism is scalable and adaptable enough to be suitable for large-scale dynamic computing environments.