LLMs are great at language but bad at logic. Ask Claude to solve SEND + MORE = MONEY and it guesses wrong. prolog-reasoner gives LLMs access to SWI-Prolog via MCP. The LLM writes Prolog, Prolog solves it, the answer comes back correct and debuggable. Results on 30 logic problems: 73% → 90% accuracy. The gain is in constraint satisfaction and multi-step reasoning — where guessing fails. MCP server for Claude Desktop, Claude Code, Cursor. Rule bases for reusable domain logic. Open source (MIT).