Franchise Hockey Manager 12 quietly has one of the best features in sports sims: Export CSV, which dumps your entire save — rosters, stats, ratings, scouting grades, ~2 million rows across ~48 tables — to flat files.

I wrapped that export in a local MCP (Model Context Protocol) server and now I run my OHL team with an AI assistant GM. The setup is a Python ingest script watches the export directory and loads each new export into SQLite as a snapshot keyed to the in-game date. The MCP server sits on top of that database and exposes tools to the AI — raw SQL queries against any table, standings, roster pulls, and a development tool that diffs scouting ratings between any two snapshots, so it catches every half-star of prospect movement between exports.

Because every answer is grounded in tool calls against my actual save, the assistant doesn't hallucinate players or ratings but it queries them. It builds draft boards before the priority selection, grades my draft class in October when teen potential settles, writes monthly report pages, and runs a custom 100-point roster-scoring rubric (a deterministic function in the server — age pyramid shape, overager quality, import slots, goalie pipeline) so I can measure my rebuild against the blueprint instead of vibing it.

The snapshot design is the killer part: the game only shows you current ratings, but the database remembers all of them. It flagged a goalie prospect's potential dropping half a star because my coach wouldn't start him.

None of this works without the sim underneath being deep enough to be worth querying and FHM12 is. If you like management sims where the spreadsheet IS the game, this one hands you the whole spreadsheet. The export turns it into a platform.

Franchise Hockey Manager 12 quietly has one of the best features in sports sims: Export CSV, which dumps your entire save — rosters, stats, ratings, scouting grades, ~2 million rows across ~48 tables — to flat files. I wrapped that export in a local MCP (Model Context Protocol) server and now I run my OHL team with an AI assistant GM. The setup is a Python ingest script watches the export directory and loads each new export into SQLite as a snapshot keyed to the in-game date. The MCP server sits on top of that database and exposes tools to the AI — raw SQL queries against any table, standings, roster pulls, and a development tool that diffs scouting ratings between any two snapshots, so it catches every half-star of prospect movement between exports. Because every answer is grounded in tool calls against my actual save, the assistant doesn't hallucinate players or ratings but it queries them. It builds draft boards before the priority selection, grades my draft class in October when teen potential settles, writes monthly report pages, and runs a custom 100-point roster-scoring rubric (a deterministic function in the server — age pyramid shape, overager quality, import slots, goalie pipeline) so I can measure my rebuild against the blueprint instead of vibing it. The snapshot design is the killer part: the game only shows you current ratings, but the database remembers all of them. It flagged a goalie prospect's potential dropping half a star because my coach wouldn't start him. None of this works without the sim underneath being deep enough to be worth querying and FHM12 is. If you like management sims where the spreadsheet IS the game, this one hands you the whole spreadsheet. The export turns it into a platform.