Agentic FC

Agentic FC is an open-source football management simulation written in Go where the player is not a human clicking menus — the player is an AI agent issuing MCP tool calls. The agent reads the world state, sets the Manager’s mindset, and watches probabilistic football decisions cascade into match events. Humans observe the result through a terminal TUI that renders ASCII match scenes and running commentary. The world engine is seeded and deterministic, so the same seed produces the same season, making it possible to isolate whether a different agent strategy — not a different random roll — changed the outcome.

The core differentiating feature is the MCP-native control loop. Instead of wrapping a simulation in a one-shot prompt, the agent operates in a continuous read-decide-observe cycle through MCP tool calls, giving it the same kind of persistent context pressure a real management decision loop would impose. This makes the project useful not just as a game but as a repeatable environment for studying how an LLM agent degrades, fixates, or adapts across a long time horizon — something a single API call cannot reveal.

The tool fits developers who want a structured, non-trivial environment for MCP agent experiments without building their own simulation engine from scratch. It breaks down when the research question requires features the current tool surface does not expose: multi-club competition, transfer markets, or deep squad statistics. The project is at an early commit count with a small community, so gaps in documentation and missing tool endpoints are filled by reading Go source, not by searching Stack Overflow.