A particularly severe attack on routing protocols in Mobile Ad hoc Networks (MANETs) is the collusion attack, in which a pair of colluding attackers prevent routes to a specific node (targeted node) from being established. This attack has devastating impact on Optimized Link State Routing (OLSR) protocol which is one of the standard routing protocols for MANETs. When there is a collusion attack in the network, the simulation results show that the Packet Delivery Ratio (PDR) drops to 0% for the targeted node. In this paper, we present an efficient protocol to prevent the collusion attack, by incorporating an information theoretic trust framework in OLSR. This protocol tries to ascertain the presence of the colluding attackers by evaluating and quantifying their trust values based on their uncertainty measures. Entropy functions have been used to represent the uncertainty involved in a node's behavior. We call the proposed protocol as Collusion Attack Prevention - OLSR (CAP-OLSR). In CAP-OLSR, the targeted node requests neighbor nodes to make recommendations on doubtful attackers based on the nature of the attack. For this purpose two new messages called Trust Request (TREQ) and Trust Reply (TREP) have been added to the OLSR protocol. Simulation results show that CAP-OLSR can efficiently detect the colluding attackers and significantly improve the PDR of the nodes in the network. Our protocol has several advantages since it does not require any time synchronization or location information and shows significant improvement in PDR under various scenarios.