PrologMCP Enhances LLM Agent Reasoning with Standardized Symbolic Delegation.

While advanced language models excel in many areas, they often struggle with deep deductive reasoning tasks, and scaling their internal reasoning capabilities can be prohibitively expensive. This research introduces PrologMCP, an open-source server designed to address this limitation by providing a standardized interface for LLM agents to interact with Prolog, a powerful symbolic logic programming language. PrologMCP exposes Prolog as a stateful tool through the Model Context Protocol (MCP), featuring a compact interface, structured error reporting, and per-session isolation. This setup facilitates a robust "translate-run-inspect-repair" loop, allowing an LLM agent to translate a problem into Prolog, execute it, receive structured feedback, and iteratively refine its approach. This delegation offloads complex inference to a dedicated solver, complementing the LLM's natural language understanding. Evaluations against leading reasoning LLMs (Claude Sonnet 4.6, GPT-4.1, o4-mini) on subsets of PARARULE-Plus demonstrated significant improvements. The formalizer agent enhanced with PrologMCP achieved near-perfect accuracy on both general and challenging logical reasoning tasks, substantially outperforming LLMs that rely solely on internal reasoning, especially on problems designed to expose their failure modes. This highlights symbolic delegation as a powerful and inspectable alternative for enhancing AI agent reasoning.