Claude Prompts for Running a Shopify Store With MCP

The best Claude prompts for running a Shopify store name a time window, a number, or an order, say what shape the answer should take, and ask Claude to confirm before it changes anything. They also need something behind them: Claude can't see your store unless it has a tool that reads it. Below are prompts grouped by the job they do, each paired with the MESA MCP tool it relies on, so you can copy the prompt and the tool together.

Every tool here is a MESA workflow that starts with the MCP trigger. You connect Claude to your MESA MCP server once, turn on the workflows you want, and each one shows up in Claude as a tool. If you haven't set that up yet, our step-by-step guide to connecting Claude to an MCP server covers it, and the setup steps are summarized below.

What makes a Shopify prompt work in Claude

Claude reads your MCP tools' names, descriptions, and parameters, then decides which one to call and what to pass in. A prompt works when it gives Claude everything a tool needs and tells it what to do with the result. Four habits cover most of it:

• Give the window or the number. "Last week," "September 1 to 30," "five or fewer," "order 1042." An order tool with start_date and end_date parameters can't guess which week you mean.
• Say what the answer should look like. "As a table with order, customer, and total," "three bullet points," "a draft reply I can paste." Claude will format the data however you ask.
• Keep the range tight. Every row a tool returns takes room in the conversation. A week of orders is quick; a year of orders may not finish.
• Ask for a check before any change. "Show me the list before you tag anything" turns a two-step job into one you can review.

The tool side matters as much as the wording. MESA's own documentation says to name and describe each workflow clearly, because that's how Claude tells similar tools apart:

"Be as explicit as possible to help AI tools determine which of your workflows to use."MESA MCP documentation

Prompts for orders and sales

Tool: a read-only workflow with start_date and end_date parameters followed by Shopify's Get List of Orders action. The template below sets it up with both parameters in place.

• "How many orders did we get last week, and what was the total revenue?"
• "List every order from the last three days that's paid but not fulfilled yet, oldest first."
• "What were our five best-selling products in October, by units sold?"
• "Compare this week's orders with last week's: count, revenue, and average order value."
• "Which orders from yesterday used a discount code, and which codes?"
• "Did anyone order more than 10 units of a single item this month? Show the order and the customer's first name."

Claude does the counting and comparing itself from the orders the tool returns, so one well-described order tool answers all of these. For month-long questions on a busy store, ask one week at a time and have Claude add them up.

Prompts for one order and customer questions

Tool: a workflow with one order_number parameter followed by Shopify's GraphQL Query action, returning the order's payment and fulfillment status, tracking numbers, and items. This is the tool support teams reach for most, because the question usually starts with a customer's email.

• "Where is order 1042? Has it shipped, and what's the tracking number?"
• "A customer says order 1187 is missing an item. What was in the order, and how many fulfillments does it have?"
• "Look up order 1203 and draft a short, friendly reply telling the customer when it shipped and how to track it."
• "Is order 1099 paid? If it's still pending, say so plainly."

The draft-a-reply prompt is where this earns its keep: Claude reads the real status and writes the answer in one step, and you paste it into your help desk. The workflow returns the customer's first name but not their address or email, which keeps personal data out of the chat unless you choose to add it.

Prompts for inventory and reordering

Tool: a workflow with a threshold parameter followed by Shopify's GraphQL Query action, returning variants at or below that stock count with their product title, variant, SKU, and quantity. The template was written for ChatGPT, but the workflow is the same in Claude: any MCP client connected to your MESA server sees it.

• "Which variants have five or fewer left? Group them by product."
• "What's sold out right now?"
• "List everything under 10 units as a table I can send to our supplier, with SKU and current count."
• "Of the products under 5 units, which ones sold the most last week?" (This one uses the low-stock tool and the order tool together.)

The quantity this tool reads is each variant's total across locations. If you stock several warehouses, build a second tool that reads one location's inventory levels and name it so Claude can tell them apart.

Prompts for customers

Tool: a read-only workflow with start_date and end_date parameters followed by Shopify's Get List of Customers action.

• "How many new customers signed up last month?"
• "List customers created this week who haven't placed an order yet."
• "Which customers from September have placed more than one order? Show their first name and order count."
• "Summarize the tags on customers who joined in the last 30 days."

Prompts that change your store

Read-only tools can't change anything, which makes them the right place to start. When you want Claude to act, give it narrow write tools, one job each, rather than broad access. Two that pair well with the tools above:

Tag orders from a chat

Tool: a workflow with order_id and tag parameters followed by Shopify's Order Add Tag action.

• "Find the unfulfilled orders from Friday and show them to me. Then tag the ones over $200 as priority."
• "Tag order 1042 as replacement-sent."
• "Every order from yesterday with a gift note should get the gift tag. List them first and wait for my OK."

Rewrite product copy, with your approval

Tool: a workflow with product_id, title, and body_html parameters that looks up the product with Retrieve Product, waits at an Approval step showing the old and new copy side by side, and only then runs Shopify's Update Product.

• "Rewrite the description for product 8123456789 in two short paragraphs, keeping every fact in the current one. Apply it when I say so."
• "Make this product title shorter and lead with what it is, not the brand."

The change waits in MESA until you approve it there. If you approve after MESA's 60-second request limit, the update still goes through, but Claude won't see the confirmation, so ask it to look the product up again.

Prompts that tell your team

Tools: a workflow that posts a Slack message, and one that sends an email, each with the recipient and message as parameters. These turn an answer into an update without copying and pasting.

• "Check what's under five units and post the list to #purchasing in Slack."
• "Summarize yesterday's orders in three lines and send it to the ops channel."
• "Email our warehouse lead the unfulfilled orders older than two days."

Prompt and tool cheat sheet

Job	Example prompt	MESA MCP tool	Changes your store?
Sales totals	"How many orders last week, and what was revenue?"	Get List of Orders by date	No
Order status	"Where is order 1042?"	GraphQL Query by order number	No
Reordering	"Which variants have five or fewer left?"	GraphQL Query by stock threshold	No
New customers	"How many customers signed up last month?"	Get List of Customers by date	No
Tagging	"Tag Friday's unfulfilled orders over $200 as priority."	Order Add Tag	Yes, one tag per call
Product copy	"Rewrite this description and apply it when I say so."	Retrieve Product, Approval, Update Product	Yes, after your approval
Team updates	"Post the low-stock list to #purchasing."	Slack or Email send	No (sends a message)

Set up Claude once

1. Build one MCP workflow in MESA. Create a workflow, choose the MCP trigger, add the parameters the tool needs with plain descriptions, add the Shopify step, and turn it on. The order tool above is a good first one.
2. Copy your server URL. In the MCP trigger, click "More fields" and add the MCP Server URL. It contains your account's key, so treat it like a password.
3. Add it to Claude. In Claude, go to Customize, then Connectors, choose Add custom connector, and paste the URL. Claude Desktop can also load it from its config file, which is the method MESA's documentation walks through; this guide to Claude Desktop's MCP setup paths covers the details and the usual fixes.
4. Start a chat and ask. Every MCP workflow you turn on later shows up as another tool in the same connector, with no reconnecting.

Custom connectors aren't limited to paid plans, but the free plan allows only one:

"Free users are limited to one custom connector."Claude Help Center

That's enough for MESA, since one connector carries every workflow you build.

Keep write prompts safe

• Approve tools one at a time. Claude asks permission before it calls a tool. Anthropic's advice is to choose "Allow always" only for a server and tool you trust to run unsupervised, which in practice means read-only tools.
• Turn off write tools you don't need. You can disable individual tools for a conversation from Claude's tools menu. Claude's Research mode can call connector tools without asking again, so Anthropic recommends turning off tools that write to other apps before using it.
• Keep each write tool narrow. A tool that only adds a tag can only add a tag. That limits what a misread prompt can do.
• Mind the limits. MESA answers each MCP request within 60 seconds, and a large response fills Claude's context window, which shows up as a "response was interrupted" message. Return only the fields a question needs and keep date ranges short.

When a prompt should become a scheduled workflow

If you type the same prompt every morning, it's a workflow, not a prompt. "What's under five units?" asked daily is better as a MESA workflow on a schedule that posts the answer to Slack, and it runs whether or not anyone remembers to ask. Claude with MCP is for the questions you couldn't predict and the one-off changes you want to make from a chat. For a wider view of where each approach fits, see our guide to MCP automation for ecommerce, and if you're weighing this against Shopify's built-in assistant, our comparison of Sidekick and MCP for Shopify admin work covers that. Note that Shopify's own Dev MCP server is a developer tool: it connects AI tools to Shopify's developer resources, not to your store's orders.

Using ChatGPT instead? The same MESA tools work there; see how to connect ChatGPT to your Shopify store with MCP.

FAQs

What are good Claude prompts for running a Shopify store?

Prompts that name a time window or a specific order, ask for a clear output format, and match a tool Claude can call. For example: "List every order from the last three days that's paid but not fulfilled," "Where is order 1042?", or "Which variants have five or fewer left?" Each one needs a MESA MCP tool that reads that data from Shopify.

Can Claude access my Shopify store?

Only through tools you give it. With MESA, each tool is a workflow that starts with the MCP trigger and runs Shopify steps. You add your MESA MCP server to Claude once as a custom connector, and Claude can call any of those workflows you've turned on.

Can Claude change my store without asking?

Claude asks before it calls a tool unless you've chosen "Allow always" for it, and it can only do what your workflows do. Read-only tools can't change anything. For edits, keep each tool narrow and add an Approval step in MESA so nothing changes until you approve it.

Does this work on the free Claude plan?

Yes. Claude's help center says free users can add one custom connector, and one MESA connector carries every MCP workflow you build.

Why did Claude say its response was interrupted?

Usually a tool returned too much data for the conversation. Narrow the date range, return fewer fields from the Shopify step, limit the number of results, or start a new chat.