We present a multi-point curvature sensor based on optical fiber specklegram measurements. Apart from the current approaches, the proposed system uses an ordinary multimode fiber excited with visible light as a reflection-type probe. Besides, this method discretizes the waveguide into segments connected by joints and assumes sequential bend events, simplifying the specklegram referencing for correlation analyses and avoiding laborious deep learning processing. Sensor characterization yielded a linear response with ∼1.3∘ resolution for single curvatures, whereas shape prediction experiments in the plane resulted in maximum errors of ∼3.5∘ and ∼5.4mm for angular and linear positioning, respectively. Furthermore, exploratory tests indicated errors <2.3∘ regarding probe curvatures in the space. This research introduces a feasible, straightforward alternative to the available shape sensors, enabling applications in medical probes and soft robotics.