Agricultural data management continues to rely heavily on traditional practices, such as manual record-keeping, spreadsheet-based tracking, and semi-structured digital forms. These methods are often labour-intensive, error-prone, and difficult for non-specialist users who may lack the technical expertise to apply query languages. To address these challenges, this paper introduces a user-friendly interface that enables natural language querying of operational datasets stored in NoSQL databases. Unlike prior NL2Mongo approaches, Agri-F2P integrates the schema validation, JSON repair, and workflow features such as auto-mated dispatch ticket generation to meet agricultural compliance re-quirements. The proposed system processes free-form English text input through a structured pipeline involving lexical normalisation, schema alignment, and controlled language generation. Evaluation was conducted on 300 ground-truthed queries, confirming high correctness in execution (87.3% success rate), accurate field alignment (93.5%), and robust data handling (96.8%). In addition, a benchmarking empirical comparative analysis has been performed to compare multiple LLM Model(s) families (GPT-3.5, GPT-4o, GPT-4.1-mini/nano, GPT-5-mini/nano, Flan-T5, DeepSeek). Results showed the most recent released GPT-5 variants as inefficient due to excessive token use, higher latency, and reduced reliability in JSON validation. Conversely, legacy models like GPT-3.5 Turbo are latency-optimal, while GPT-4.1-nano is the most cost-efficient. These findings demonstrate that newer models do not necessarily outperform earlier ones in structured query generation.