Inverters with grid forming capability are becoming increasingly important in modern power distribution systems. Finite Control Set Model Predictive Control (FCS-MPC) has emerged as an effective solution for inverter system control, offering benefits such as multi-variable regulation, rapid dynamic response, and an intuitive design framework. However, the effectiveness of MPC is closely dependent on the accuracy of the underlying mathematical prediction model. To address the challenges of modelling error and uncertainty in predictive control systems, model-free strategies have emerged as a promising solution where the ultralocal model technique is one of the most utilized. In this scenario, a Model-Free Predictive Control (MF-PC) based on the ultralocal model and Extended State Observer (ESO) can effectively reduce model uncertainties. This article proposes an ESO-based ultralocal model for controlling the voltage of a Two-Level Voltage Source Inverter (2L-VSI) equipped with an LCL output filter. Hardware-in-the-Loop (HIL) experiments are used to verify the real-time feasibility and applicability of the proposed MF-PC.