We present an algorithm that plans a sequence of multi-contacts stances in order to solve the locomotion and manipulation planning problems for humanoid robots and equivalently for dexterous hands. The algorithm uses a Best-First planning approach to explore the stances space and relies on an inverse statics stance solver to generate postures in the configuration space of the system realizing a given stance. Results show that locomotion and manipulation motion planning problems can be solved within the same framework in a unified way.