My store gives me more numbers than I can use at once. I ask AI to find the few that deserve a closer look, then I make the call.
I build Hoard and operate Amor Honor Gaming, a TCGplayer store. So I have a reason to be clear about what the assistant can do and what I still have to check myself. The prompts here show how I approach the work; they aren't transcripts from my store or evidence of a measured result.
I start with one question: “What would I need to see before making this decision?” That might mean the last successful sync, the rule behind a strange price, or the orders in today's pull. The assistant is good at gathering those pieces. It doesn't get to fill in a missing number or quietly change a listing.
I use two connections for different jobs. tcgmcp is free and public: card market-price lookups, known printings, decklist estimates, and market trends. It knows nothing about my inventory or orders. Hoard MCP connects to my Hoard account, so an authorized assistant can read my TCGplayer inventory, rules, sales, customers, orders, and sync status. Some actions can also change data, with different safeguards depending on the action. There are setup guides for Claude, ChatGPT, and Codex.
If I'm pricing a deck someone pasted into chat, I can start with the public decklist workflow. If I'm asking why one of my cards is listed at a certain price, I need the private store connection. A public market price alone cannot explain my floor, rule, lock, or last sync.
Before I act on a recent order or a pricing gap, I ask: “Is my Hoard sync healthy? When did it last finish, and is one running now? Don't trigger a sync.” That is a read. If the answer says the desktop agent is offline or the store data is stale, I fix that before trusting a list of “today's” opportunities. The sync-health guide shows what the verdict means; triggering a new sync is a separate action.
A confident answer on old inventory is still an old answer. “Last sync failed” is useful information; a ranking that hides that fact isn't.
For pricing, I ask three read-only questions in one conversation: “Which of my cards are below market?”, “What suggestions should I review?”, and “What were my biggest movers in the last seven days?” That gives me a short list to investigate instead of another screen to scan. The weekly pricing-check walkthrough gives the sequence; the underpriced-inventory guide keeps my own listings separate from cards I might buy.
A gap against market is a lead, not an instruction to raise the price. Condition, finish, source, my rule, and a price lock can all change the right action. When the pattern is broader than one card, I ask for a pricing-rule proposal that I can review in Hoard before approving it.
When several cards point to the same rule, I audit the rule against my actual inventory before asking for a new multiplier.
For a suspicious listing, my follow-up is: “Why is this card priced this way? Show the governing rule, the source price, any floor or cap, and any blocker. Don't change it.” Hoard can explain the stored price and evaluate the current rule for a card. If those differ, that's a reason to investigate sync or rule state, not a reason to trust whichever number sounds better. Here's the card-price explanation guide, including what to check when the answer is incomplete.
I don't ask a chat client to invent a one-off manual override for a single listing. If a rule is wrong, I want the assistant to prepare the rule change and show the affected cards and impact. Hoard's capability map separates those reviewed rule proposals from other actions that can apply directly.
A sales question I can hand off: “Compare the last three covered months. Show gross sales, fees, refunds, and order count, and tell me which months are missing.” Covered matters. Hoard can only summarize the reports it has; a missing month isn't a zero. The sales recap guide explains the limits.
For customers, I can ask who bought more than once or whether a small group drives a lot of revenue. Then I open the buyer in Hoard. The assistant can surface order history and existing context, but my private 1–5-star rating of a customer is mine to set in the customer panel. It isn't TCGplayer feedback and isn't posted to the buyer. The repeat-customer guide connects the read to that human judgment.
For an order day, I can ask: “Start today's pull session and tell me the eligible order count by game.” Unlike the earlier questions, starting a pull is a direct write to Hoard: it snapshots eligible open orders. The assistant can check progress later, but I still print from the dashboard, pick and inspect the physical cards, and handle exceptions on the tablet. The AI pull-session guide covers that boundary, and the full pull-sheet workflow shows the physical process.
I don't delegate card condition, whether a card is actually on the shelf, or the final check before an order ships. I don't treat a public market quote as the price my store should charge. I review any proposed pricing-rule change before it applies, and I read the details of a Smart Action before accepting it because some actions apply directly. AI is most valuable to me when it makes these choices more legible, not when it pretends the choices disappeared.
These examples are for a TCGplayer store connected to Hoard; they aren't a claim that Hoard can operate every seller platform, message buyers, or perform physical fulfillment. If you want the full task list and current read/write boundaries, start with the AI workflow hub and the capability map.
Connect Hoard to the assistant you already use, check sync health, and start with a read-only pricing or sales question. You can explore public card prices with tcgmcp before connecting a store.
Connect an assistant to your store