Recommender systems play an important role in providing items to users that align with their goals. However, explaining the relevance of recommended items remains challenging; a particularly important task for high-stakes domains like agriculture where both suitability and interpretability of recommendations is crucial. Knowledge Graph (KG)-based approaches can provide suitable recommendations, but lack fluent explanations. Recent Large Language Model (LLM)-based recommender systems excel in natural explanations, yet are limited in factuality or coherence. Combining strengths of LLM-based recommendations with knowledge presents a promising direction. We introduce KRAGGER (Knowledge Retrieval-Augmented Generation for Graph-based Explainable Recommendation), a composite framework combining KG- and LLM-based methods. We apply KRAGGER to an agricultural use case by testing its ability to recommend technologies to farmers while explaining provided matches. The contributions of this paper are: 1) KRAGGER’s three recommendation methods: Graph-Graph, Graph-Attribute and Attribute-Attribute matching, and 2) KRAGGER’s three explanation methods: Rule-Based (RB), LLM and RB-LLM, and 3) an agricultural recommendation KG dataset describing farmers and farm technologies. Recommendation performance of KRAGGER is measured quantitatively with catalog coverage and intra-list diversity on five different embedding models. Additionally, a user study is performed with domain experts to validate recommendation performance and measure simplicity, fluency, coherence and trustworthiness of explanations. Results show the fine-grained attribute matching recommendation performs best, which forms the basis of the highest rated RB-LLM rephrasing explanation method. Thus, KRAGGER shows promising capabilities for grounded, yet natural recommendation explanations, underscoring the importance of composite AI (KGs and LLMs) to support difficult decision making.