Why MCP changes how a dispatch board works

For two decades, dispatch boards have been built around the same loop: a person drags a card, the board recalculates, the person decides what's next. The board is honest with humans. It's silent with AI.

That's because the AI didn't have hands.

What "read-only" looked like

Before MCP, the best you could do with an AI assistant and a dispatch tool was screen scraping and prompt engineering. You'd paste a screenshot or a CSV export. The AI would read it, give you back text. You'd squint at the text and try to apply it manually. The loop was: human reads board → human prompts AI → AI suggests in prose → human translates back to board.

Each step lost fidelity. Each step took attention. None of it was real-time.

What MCP gives the AI

Three things, in order:

1. A standardized tool interface. The same MCP server can serve Claude, ChatGPT, Cursor, or any client that adopts the spec. No per-vendor integration.
2. Direct, authenticated access. OAuth 2.1 with PKCE. The user authorizes once; the assistant has scoped access until it's revoked.
3. A live working surface. Tools return real data. Tools that mutate change state. The AI isn't drafting suggestions; it's operating the system.

The shift is from "AI as advisor" to "AI as operator."

What this does to the board

The board doesn't go away. It becomes the consensus view between you and the AI. You see the same cards Claude sees. When Claude moves something, you see it move. When you drag something, Claude can find out.

The "AI as operator" pattern works because the human can step in any time. Drag-and-drop on the board is faster than chat for spatial edits. Chat is faster for bulk operations and conditional logic ("move everything outdoor on Friday because of the storm").

You use whichever fits the moment.

What it doesn't do

MCP doesn't make the AI smarter. It doesn't know your customers better than you do. It will hallucinate appointment details if you don't ask it to look them up first. The right pattern is to tell the assistant: "Use appointments-list before saying anything specific." It will, and the hallucination problem mostly goes away.

It also doesn't replace decisions. Whether to push Friday's work to Monday because of a storm is your call. The assistant will execute the call cleanly once you make it.

The wedge

Calendesk shipped AI booking. Jobber shipped the dispatch board. Both are good at what they do, neither is both. The wedge for Jasning is: the dispatch board your AI can actually read and write. That's the whole thing.

Try the docs for the setup. Five minutes for OAuth, then ask Claude to plan your week.