LLM-Driven Smart Agents for Surface Logging: Enhancing Drilling and Geological Intelligence with MCP and A2A Protocols

This paper presents the design and evaluation of LLM-driven Smart Agents for surface logging workflows in drilling operations, integrating two key industry protocols: Model Context Protocol (MCP) and Agent-to-Agent (A2A) communication. The MCP-based architecture enabled an LLM to autonomously manage well metadata, build datasets, and train a lithology classification model with 81% accuracy. Separately, the A2A-based framework expanded on previous multi-agent research by coordinating distributed agents—each running in its own environment and built with different frameworks—for a drilling safety assessment task involving Mechanical Specific Energy and Drilling Strength anomaly detection and cavings image analysis. While each architecture was evaluated independently, both demonstrate complementary strengths: MCP offers centralized control for tool orchestration, while A2A enables modular, scalable, and framework-agnostic multi-agent coordination. Together, they provide a flexible and interoperable foundation for future AI-driven surface logging operations, paving the way for more autonomous, collaborative, and production-ready digital drilling workflows.