This paper is concerned with the design of online data-driven model predictive control (MPC) for switched linear systems with online-generated subsystems and unknowns witching signals. The online input-state data are used to detect switching instants, and update the MPC law directly to match the new subsystem bypassing identifying its model. Once a switching is detected, a fast-to slow excitation scheme will be executed until a suitable MPC law is attainable. Specifically, the fast-sampling exci