What to ask your AI agent once it is connected to InfiPlex: prompts by role, with the tools behind each
Connecting Claude, ChatGPT via Codex, or any MCP client to InfiPlex takes five minutes. What you ask it in the first week decides whether it becomes the way your team runs orders or a novelty. These are the prompts that stick: grouped by who asks them, with the exact InfiPlex MCP tools the agent calls and what you should expect back. Every chain here is real and runs against the live server; copy a prompt, change the ids and SKUs to yours, and go.
Two counted searches per channel: created today, and created today with fully_shipped 0. The agent reads order_source values from the first result so channel names match exactly.
Show me every order over $500 from the last 7 days that has not shipped.
If you give a marketplace order number instead, the agent resolves it with order_search text_search first, then order_get returns lines, addresses, totals, and shipment state.
Why did order 4418211 ship from Reno when the customer is in Ohio?
Returns the routing event log written when the order was allocated: eligible warehouses, zip rule checks, on-hand at that moment, units taken per warehouse. See the geo routing guide for the rules themselves.
What happened with the Temu orders yesterday? Anything fail?
Finds the Temu log types, searches the failed type for yesterday, and summarizes each failure with its log_message. Pair with order_get on any order id in the message.
Add a note to order 4418211 that the customer called about delivery.
inventory_get without a warehouse gives the total; with warehouse_id gives one building. The agent resolves warehouse names to ids with warehouse_search.
HB-2201 shows 3 on hand. Walk me through how it got there.
Newest first: each change with the resulting total, whether it was a delta or an absolute set, and the cause, for example an order number or a product creation.
Record UPS Ground tracking 1Z999AA10123456784 against order 4418211, all lines in one box, 48 oz.
The agent reads order_get for the line itemid values (package_items is keyed by itemid, not SKU), then previews the shipment. You confirm, and tracking is submitted to the channel.
One call, one preview listing every row as quantity_to_set. A row whose SKU cannot be resolved comes back null rather than failing the batch, and the agent reports those rows.
Put HB-2201 in bin A-14 at Reno and bin C-02 at Columbus.
Same weighted reorder point the Restock Alerts email uses. Rows include reorder_point, total_inventory (negative when oversold), lead time, and safety stock. Ask for a count first on large catalogs.
Checks each SKU does not already exist, then one previewed batch create. A duplicate SKU would fragment inventory and pricing across channels, so the check is deliberate.
Reads 27.50, previews the change to 24.99 with the server's Tool / Effect / Arguments block and a 10 minute token. You confirm. Goes to every connected channel.
The agent finds the SKUs, reads each price, and previews each change one at a time; there is no bulk price tool, so you approve each SKU. Say so if you want it to stop after the first.
Settings and channel health
Settings and channel health
Discover what a setting is called, read it, and, on a key that allows it, change it.
Returns every setting whose name contains walmart, with its current value and a description of what it controls. This is how the agent learns setting names; it never guesses them.
Is Amazon order pulling on, and how often does it run?
Searches Amazon Get Orders: Failed for the window, reads the error, confirms pulling is enabled and that the Selling Partner authorization is set, and if the failures are Unauthorized offers a fresh connect link for the Seller Central owner.
Change the restock alert email to ops@ourcompany.com.
shopify_1_pull_orders_enabled matches the arming pattern, so this additionally requires MCP Allow Arming. Without it the server refuses and the agent tells you which flag is missing.
3PL and multi-client
3PL and multi-client
Onboarding client accounts and answering for many brands from one place.
We signed Acme Outdoors. Give me a link so they can connect their Walmart account.
Returns a one time authorization URL for eal__walmart_connect. Hand it to whoever controls the Walmart Seller Center login; they authorize with no InfiPlex account. Expires the same day.
Create a warehouse record for our new Dallas facility, zip rules on.
Writes the additional info field the NetSuite push consumes. The name must match one the account already uses; the agent confirms the exact name rather than guessing. See the field mapping guide.
Address, identifier, zip rule flag, fulfillment ordering. Change any of it with warehouse_set, previewed.
The first week
A five-day plan that makes it stick
Start on a read-only key. Day one: the morning count, one order lookup, one routing explanation. Day two: the restock list and a change history on whatever surprised you. Day three: a log search on the channel that gives you the most trouble. Day four: add Write Orders and Write Inventory, record one real shipment and one receipt through the agent, and watch the previews. Day five: ask the question you would normally open three tabs to answer. If the agent answers it in one message, you have your new interface. The write safety guide explains what each key setting unlocks as you widen the key.
Why this works
Headless order management, one question at a time
None of these prompts are scripted responses. Each one is the agent reading the InfiPlex MCP server's tool list, choosing tools, calling your instance, and composing an answer from live data. That is what connecting your AI agents to InfiPlex means in practice: the platform keeps pulling orders from 130+ channels, routing across warehouses, syncing inventory, and pushing to NetSuite, QuickBooks Online, and your 3PL, and the screens become optional. New prompts work the moment you think of them, and new tools appear in the agent the moment they ship, with nothing to install.
No. The agent maps natural language to tools using each tool's description and schema. These are examples of the questions that map cleanly. Vague asks ("fix inventory") get a clarifying question; specific asks ("receive 200 of HB-2201 into Reno") get a preview.
Every prompt whose chain contains only read tools: searches, gets, routing explanation, change history, restock list, price reads, setting reads and searches, log searches, and connect links. That is most of the operations, purchasing, and channel health sections.
Only one SKU per preview today, because price_set is a single-SKU tool. The agent will walk a list and ask you to confirm each. Inventory does have a bulk tool, inventory_set_bulk, with one preview for the whole batch.
For reads, you get an answer to a slightly different question and ask again. For writes, the preview names the tool and the exact arguments before anything happens, so a wrong tool is visible and you decline. Naming the tool yourself ("use order_tracking_create") also works.
Yes. Issue them a read-only key, connect their assistant, and send them this page. They can answer their own "where is my order" and "what is low" questions without an InfiPlex login or your time.
Order ids, SKUs, and warehouse names in the prompts are examples. Tool chains reflect the InfiPlex MCP server as deployed September 2026. Claude is a trademark of Anthropic, PBC. ChatGPT and Codex are trademarks of OpenAI. Model Context Protocol is an open standard.
Our website uses cookies to help provide the best user experience. By continuing to use this site, you consent to the use of cookies as described in our Privacy Policy. OK