The introduction of AI-based solutions, especially those based on large language models (LLM) as autonomous agents, to Supply Chain Management (SCM) and logistics may lead to improving the decision support. Nevertheless, a lot of potentially valuable predictive capabilities exist in well-known statistical algorithms (e.g., regressions and time-series models) with which members of the logistics community are familiar and comfortable performing tasks such as demand forecasting or transit time estimation. In this research, a new way of how these paradigms can be united is suggested and examined in terms of encapsulating traditional machine learning prediction algorithms as services that can be called through the Model Context Protocol (MCP). We show how to use the MCP server to expose an Exponential Smoothing model, allowing LLM-based agents to invoke it for quantitative tasks, making it a statistically supported, robust solution. Such an approach can empower LLMs in terms of their reasoning to be enhanced with established quantitative knowledge, making their decision support recommendations more reliable and accurate to be implemented in dynamic logistics conditions. The architectural pattern and the possible advantages in developing more credible and effective AI-augmented SCM systems are outlined, which makes the way towards increased synergy and collaboration with traditional data science and the emergent agentic AI.