Your agent sets it up. Kaselles computes. You decide.

Start with days, periods, teachers, groups and rooms. Ask the agent to check weekly teaching loads and availability before generating. Missing information must be requested or flagged: a list of subjects does not tell the agent who teaches them.

Local actions can configure resources, rules and preferences, validate the project and request generation. Changes appear in the application. Pay particular attention to operations that replace whole lists, such as the grid or a resource's availability.

A chatbot response is not, by itself, proof of validity. OR-Tools CP-SAT searches for solutions satisfying the model's hard constraints while weighing preferences. Reaching a time limit does not prove a problem impossible.

MCP is the technical connection, but you do not need to learn tool names to describe your school. You do need to check the client, permissions and shared data. Save a copy before substantial changes and review the timetable before publishing it.