We consider the problem of deploying a team of agents (robots) for the foraging problem. In this problem agents have to collect disseminated resources in an unknown environment. They must therefore be endowed with exploration and path-planning abilities. This paper presents a reactive multiagent system that is able to simultaneously perform the two desired activities~ - exploration and path-planning - in unknown and complex environments. To develop this multiagent system, we have designed a distributed and asynchronous version of Barraquand's algorithm that builds an optimal Artificial Potential Field (APF). Our algorithm relies on agents with very limited perceptions that only mark their environment with integer values. The algorithm does not require any costly mechanism to be present in the environment to manage dynamic phenomena such as evaporation or propagation. We show that the APF built by our algorithm converges to optimal paths. The model is extended to deal with the multi-sources foraging problem. Simulations show that it is more time-efficient than the standard pheromone-based ant algorithm. Moreover, our approach is also able to address the problem in any kind of environment such as mazes.