We propose a information theoretical framework to capture transition and information costs of network navigation models. Based on the minimum description length principle and the Markov decision process, we demonstrate that efficient global navigation can be achieved with only partial information. Additionally, we derived a scalable algorithm for optimal solutions under certain conditions. The proposed algorithm can be interpreted as a dynamical process on network, making it a useful tool for analysing and understanding navigation strategies on real world networks.