The Object Constraint Language (OCL) is a declar- ative language that provides constraint and object query expres- sions on any MOF model. OCL can be used to add precision and conciseness to the UML models. Despite its advantages, the unfamiliar syntax of OCL contributed to its lower adoption by software practitioners. This paper experiments with prompt engineering and Large Language Models (LLM) for OCL con- straint generation from natural language. LLMs, such as GPT-3, have achieved substantial gains in many NLP tasks, including text generation and semantic parsing. Similarly, researchers have improved on downstream tasks by fine-tuning the LLMs for the target task. Codex, a GPT-3 descendant by OpenAI, is fine-tuned on publicly available code from GitHub and has proven the ability to generate code in many programming languages, powering the AI-pair programmer Copilot. One way to exploit Codex is to engineer prompts for the downstream task. In this paper, we investigate the reliability of the OCL constraints generated by Codex, given the specification in natural language. To accomplish this, we collected a dataset of 15 UML models with 169 specifica- tions and followed a prompt engineering approach. We manually crafted a template with slots to fill in the UML information and task description, following the prefix shape to complete the template with the generated OCL constraint. Both zero- and few-shot learning methods were adopted in the experiments. The evaluation is reported by measuring the syntactic validity and the execution accuracy scores of the generated OCL constraints. Our findings suggest that by enriching the prompts with the UML information of the model and with influence from the correct behaviour of the few-shot examples, the reliability of the generated OCL constraints increases. In addition, we also investigated their similarity by calculating the cosine metric between the correctly generated OCL constraints and their corresponding human-written ground truth. The results indicate that, on average, the generated OCL constraints were similar to their ground truth.