Poultry diets are routinely under- or over-formulated. Precision feeding aims to address this inefficiency. Miniaturized near-infrared spectrometers (micro-NIRS) may provide the necessary accurate and timely knowledge on the nutritional composition of feed ingredients and diets. In this study, we present the Pocket NIR, a novel micro-NIRS based on a unique optical arrangement of two complementary MEMS (micro-electromechanical system) Fabry-Pérot interferometers oriented toward the same focal spot. We developed 30 partial least squares regression models to demonstrate its analytical potential for rapid on-site analysis of protein, fat, fiber, water-soluble carbohydrates (WSC), moisture, and ash in poultry feed, corn, wheat, soybean, and DDGS (distillers’ dried grains with solubles). A library of 1437 reference samples, 248–358 samples per material, was used for calibration and hold-out validation of these models. Cross-validation was used to select the best spectra pre-processing steps from the commonly used methods for transformation, scatter correction, smoothing, and differentiation of NIRS spectra. With exceptions, models for protein, fat, and moisture had a ratio of performance to deviation (RPD) larger than 3 on a held-out dataset. Model performances with an RPD > 2 were observed for the majority of models for fiber, WSC, and ash. The heterogeneity of the material and the variability of the nutrient parameters co-determined the models’ performance. With its mobile app and cloud-based backend, the Pocket NIR could in the future assist precision feeding by offering nutritional advice and diet formulation suggestions based on its data, and considering ingredient availability, costs, traceability of supplier, and production management.