From the Volleyland: a multiplayer learning game architecture map
Agents get strict, self-describing tools that are checked twice
strength · medium · verified
Each MCP tool is an operation with a zod input schema that carries field titles and descriptions from contracts. The MCP SDK validates arguments, then OperationRunner validates again with refinements. Refusals return as tool results a model can act on. World writes and import_questions take dryRun, every write names expectedVersion, and destructive operations carry destructiveHint. These are the schema and correction loops an agent needs.
Schema discipline is the strongest of the seven concerns here; identity, provenance and the human escape hatch are the gaps.
The trail
Open on the mapMore from this board
- One misnamed MCP operation can stop the whole game server at startup
- An agent's change cannot be told apart from an admin's or reconstructed later
- Production deploys start closed: token required, MCP off, answer keys withheld from readers
- The admin API is open to everyone unless NODE_ENV is exactly production