Scientific workflow is the established method to manage models and data analyses in computational materials science consisting of many interrelated steps (jobs) running on different computing resources. This approach improves the reproducibility of such models, as well as the provenance and reusability of the data from all steps. However, the practical usability and productivity of this approach are not satisfactory and, therefore, the willingness to adopt workflow management systems in practice is still limited. We suggest a solution for this issue pursuing a model-driven engineering strategy, which includes a domain specific language and a platform to make the workflow management and the workload management systems fully transparent. On the particular example of modeling of catalysts for oxygen reduction reaction, we compare critically different aspects of two approaches: based on a traditional bare workflow management system and on the newly proposed textM language, and its supporting tools. We find that the proposed approach introduces substantial improvements and benefits over the traditional one and anticipate production deployment of textM, for example in virtual materials design projects.