Dev hacks C# MCP server and microsoft agents for LLM BattleTech bot—ends up with random moves that guzzle a million tokens

A developer has created a Large Language Model (LLM) powered bot to play the game BattleTech, using a.NET application called MakaMek. The bot utilizes the Microsoft Agent Framework and an MCP server to query the game state and make decisions. The developer implemented the MCP server using C# and the ModelContextProtocol SDK, allowing agents to access the game state remotely. The bot uses a local LLM model, such as qwen3-vl-8b, or a frontier model like GPT-5.2, to make decisions, but the results have been mixed, with the bot's actions often feeling random and not making sense. Despite this, the project has provided a valuable learning experience and a playground for further improvement, with potential areas for refinement including optimizing prompts and introducing new tool types, such as RAG, to provide relevant BattleTech rules. The project demonstrates the challenges of applying AI to complex games like BattleTech.