I told Claude to plan my week. It did.

The pitch is "plan dispatch by talking to it." The fastest way to know if that's true is to try.

Monday morning. Three drivers. Eight jobs in the queue. I open Claude.

Me: Plan tomorrow. Cluster anything in Brussels for Eva, anything in Antwerp for Kai, the rest on Tom.

Claude reads the queue through the Jasning MCP server, looks at driver regions, and proposes a plan. Two cards land on Eva's lane. Three on Kai's. The remaining three on Tom. The board updates while I read the reply.

Me: Tom can't start before 09:30 tomorrow.

Claude shifts Tom's first stop. Workload bar nudges down. Done.

That's it. The whole exchange was ninety seconds.

What it's actually doing

The Model Context Protocol gives Claude two things at once: a list of tools (appointments-list, availability-suggest, appointments-create, etc.) and the freedom to chain them. So when I say "plan tomorrow," Claude:

1. Calls appointments-list for the next day to see what's already scheduled
2. Calls queue-list to see what needs slotting
3. Calls drivers-list to know who's available and where they usually work
4. Calls availability-suggest to find open windows
5. Calls appointments-create for each new booking, with from_queue_item_id so the queue items drop atomically

I never see those calls. The interface is plain English in, plain English out, with the board updating in the background.

What broke (and what didn't)

The first time I tried this I got over-eager prompts. "Plan my whole week." Claude tried, ran out of context juggling fifty appointments, and got confused. The fix wasn't a model upgrade. It was scoping: "plan tomorrow," then "now Wednesday," then "now Thursday." One day at a time. The board is the working memory between turns.

The other surprise: Claude is happy to ask. When I gave it ambiguous instructions ("schedule the Verlinden job"), it asked which Verlinden (there were three) instead of guessing.

The board still matters

I tried to do everything by chat for a week. It was fine for booking and rescheduling. It was slow for "move this card to that lane." Drag-and-drop is still faster for spatial edits, and the board is still the place I look first thing in the morning.

That's the deal Jasning makes: the board is the truth, the AI is a hands-free way to manipulate it.

Try it yourself

If you have a Claude subscription and a free Jasning beta account, the setup is five minutes. The docs walk through it.

If you don't have either: start with the board, get a day or two of dispatching under your belt, then add the AI. Both surfaces work on the same data, so nothing you do is wasted.