Leveraging the path diversity in heterogeneous wireless networks by Multi-path TCP (MPTCP) not only depends on end-user's decisions but also on other competitors who are looking for maximizing their benefits. The incomplete information repeated Bayesian game is proposed to enhance the MPTCP throughput in a resource-shared wireless network context. Mobile nodes establish the initial path in the first stage by choosing the best opening sub-flows connection. Moreover, by receiving feedbacks regarding the preference of other opponents the repeated Bayesian game in the second stage improves the achievable throughput via selecting the best combination of networks. The numerical and simulation results demonstrate that the proposed algorithm could at least achieve (17%-24%) more throughput than WiFi in the initial path selection and (11%-14%) more throughput than MPTCP.