Clearing the table▍
Clearing the table▍
Guide · 4 min
A tarot MCP is a small server that, through the Model Context Protocol, hands an AI a real tarot deck to draw from. Instead of improvising cards out of words, the assistant calls a tool that shuffles seventy-eight, turns one upright or reversed, and gives back the card that actually came up. Familiar is that server.
MCP — the Model Context Protocol — is an open standard for connecting AI assistants to outside tools and data. A model can read its training and your message, but it cannot shuffle a deck or settle a coin toss on its own. MCP is the bridge: it lets Claude, ChatGPT and the others reach a tool that can. A tarot MCP server is one of those tools, and the thing it holds is a deck.
Connected, your AI gains a small set of deck tools: it can shuffle, cut, draw a single card, lay a named spread, or browse a deck face by face. Every draw is a real shuffle of all seventy-eight — even odds, each card a true one-in-seventy-eight, upright or reversed at equal chance — and it returns the card’s art and name, not a guess. The reading that follows is your assistant’s; the cards are the deck’s.
One server answers every MCP client. Add it once and read in whichever AI you already talk to — the connector is identical, only the adding steps differ.
Ask a chat model for a card and it will give you one — fluent, plausible, and chosen the way the next word is chosen, sometimes by what you seem to want to hear. A tarot MCP takes that choice away from the model. The card is drawn before it is read, so what you sit with is what the deck turned up, not what the assistant reached for. For some that is a detail; for anyone who values that a card came up on its own, it is the whole point.
New to the cards themselves? Every one of the seventy-eight has a full meaning to read, in two traditions.