How to Connect Your Shopify Store Data to ChatGPT or Claude (2026)

You can connect your Shopify store data to ChatGPT or Claude in five ways, from a CSV upload to your own MCP server. This guide ranks them from simplest to most technical and checks each one against the official documentation: Shopify's documentation on custom apps and Admin API access scopes for orders, products, customers and analytics, OpenAI's documentation on ChatGPT connectors and custom MCP servers, and Anthropic's documentation on Claude custom connectors and remote MCP. Everything was checked on November 3, 2026. Menu names in both AI apps change often, so treat the paths below as a snapshot of that date.

If your real question is "where am I losing sales?", read the section on order data versus session data before you pick. It decides which options can answer it.

What MCP is, in one paragraph

The Model Context Protocol (MCP) is an open standard for connecting AI applications to outside systems. Anthropic open-sourced it in November 2024, and the official MCP site compares it to a USB-C port: one standard plug between AI apps and data sources, tools and workflows. An MCP server exposes tools, such as "get sessions by landing page." An MCP client (ChatGPT, Claude, Claude Desktop, Cursor and many others) calls those tools when your question needs them. As of this writing, ChatGPT lists these connections under Plugins and Claude under Connectors.

The five options compared

Option 1: Export reports and upload them

The no-setup route. In your Shopify admin, go to Analytics > Reports, open a report, click the horizontal menu icon, then Export. Shopify exports CSV, XML, JSONL or Parquet. Upload the file to ChatGPT or Claude and ask your questions, or import it into Google Sheets to keep a running history.

For lost-sales questions, export the behavior reports (Sessions by landing page, Conversion rate breakdown, Sessions by device) and the acquisition reports such as Sessions by referrer, not only sales.

Limits: it's a snapshot, so the answer goes stale with the next day's traffic. The AI can only answer from the columns you exported, and every follow-up that needs a different cut means another export.

Option 2: Shopify's own AI tools

If you don't need ChatGPT or Claude specifically, Sidekick, the AI assistant inside your Shopify admin, can answer questions about acquisition data such as sessions by referrer and write ShopifyQL queries for custom reports.

The official Shopify plugin for ChatGPT and connector for Claude

Shopify publishes its own integrations for ChatGPT, Claude and other AI tools, including Gemini, Grok, Perplexity and Microsoft Copilot. You install them from the AI app, not the Shopify App Store:

1. On a desktop browser, open the Shopify plugin for ChatGPT or the Shopify connector for Claude.
2. Connect, sign in to your store, and review the access requested in your Shopify admin.
3. Approve read-only access if you only want answers. Shopify's authorization guide says a connected tool can read but not change data unless you approve write access, and it can never do more than the staff account it acts through.

According to Shopify's pages for the ChatGPT plugin and the Claude connector, they read orders without changing them (up to 50 at a time), look up customers, edit products and inventory if you allow writes, and answer analytics questions about sales, orders, products, customers and fulfillment as tables and charts, such as "How many new versus returning customers did I get this month?" Shopify advises one short question at a time.

Shopify's own advice is to use the official integration and not to create an API access token or build your own. For sales questions, that's the right default. The gap for lost-sales questions: neither help page lists sessions, landing pages or traffic sources among the analytics covered. Test it by asking for conversion rate by landing page.

Shopify's other MCP servers serve different jobs. The Dev MCP server gives coding tools Shopify's docs and API schemas, not store data. The Shopify CLI connector lets local developer tools such as Claude Code, Codex and Cursor work with store data through the API. The Customer Accounts MCP and the Universal Commerce Protocol servers, which replaced Storefront MCP's catalog and cart tools, serve AI shopping agents, not merchant analytics.

Option 3: Zapier or Make

Zapier MCP connects Claude, ChatGPT, Cursor and other MCP clients to apps in your Zapier account, and you choose which actions the AI can use. Its Shopify MCP actions include searches such as Find Order, Find Customers and Find Products by Title, plus write actions. Per Zapier's docs, each successful tool call uses two tasks from your plan. Make's MCP server lets AI clients run your Make scenarios as tools, connected by OAuth or an API key.

These shine for operations, such as "find order 1234 and tag the customer," or a Zap on the New Order trigger that logs orders to a Google Sheet. They're weak for analysis: Zapier's Shopify integration lists order, customer, product and inventory triggers and searches, but no report or session data.

Option 4: A custom app with Admin API access and your own MCP server

The developer route, for when you want specific tools (such as a ShopifyQL query runner) and someone to maintain them.

1. Create the app. Shopify custom apps are now created in the Dev Dashboard; custom apps made in the admin before January 1, 2026 are legacy apps. You need the App developer role. In the Dev Dashboard, go to Apps > Create app > Start from Dev Dashboard, pick your scopes on the Versions page, click Release, then install the app on your store.

2. Choose read-only scopes. From Shopify's access scopes reference:

The shopifyqlQuery endpoint also requires level 2 protected customer data access, which Shopify says is available to custom apps.

3. Get a token. In App settings, copy the client ID and secret. Apps owned by your own organization can use the client credentials grant to request an access token; tokens are valid for 24 hours, so your server requests new ones as needed.

4. Build the MCP server. Keep the credentials on the server and expose a few read-only tools. With read_reports, one tool can run ShopifyQL against the sessions dataset, which has one row per session:

FROM sessions
SHOW sessions, sessions_with_cart_additions, sessions_that_completed_checkout, conversion_rate
WHERE human_or_bot_session = 'human'
GROUP BY landing_page_path
SINCE -30d
ORDER BY sessions DESC
LIMIT 20

5. Connect it. Host it at a public HTTPS URL to use it as a remote connector in ChatGPT or Claude (steps below), or run it locally for Claude Desktop.

Option 5: An analytics app that ships its own MCP server

Some analytics apps run an MCP server for their own data, so you skip the build. Check four things: whether it records sessions of shoppers who didn't buy, whether it links repeat visits into one journey, which sources it joins (Shopify, GA4, ad platforms), and whether its tools are read-only or can change things.

Ariven, which we make, is one example. From install it records visits into its own database, about 140 parameters per visit with no custom event setup, and links a shopper's repeat visits into one journey, so multi-session paths aren't reduced to last-touch. You can connect Shopify, GA4, Google Ads and Meta to enrich it. Its MCP server connects that data to Claude, ChatGPT, Codex or any MCP client, and it's included on every plan, from $99/month billed through Shopify with a 30-day free trial. Setup is installing the app and connecting Ariven's MCP to your AI client.

How to add a custom MCP connector in ChatGPT

Per OpenAI's custom MCP server guide, as of November 3, 2026:

1. Open ChatGPT on the web and go to Plugins.
2. Select the plus button, then Add custom MCP server.
3. Enter a name and an optional description.
4. Under Connection, enter the server URL, including its MCP path.
5. Choose authentication: OAuth, No authentication, or OAuth or no authentication.
6. Review the risk warning, select I understand and want to continue, then Create as a plugin.
7. In a new chat, type @ and select the plugin.

ChatGPT supports SSE and streaming HTTP servers. Tools that aren't marked read-only are treated as write actions, which require your confirmation by default. Availability depends on your plan and workspace: in a Business workspace, OpenAI's help center says only admins and owners can create and publish custom MCP plugins. If you don't see Add custom MCP server, your plan or role doesn't allow it.

How to add a custom connector in Claude

Per Anthropic's custom connector guide, custom connectors work on the Free, Pro, Max, Ultra, Team and Enterprise plans; Free is limited to one.

1. Go to Customize > Connectors.
2. Click + Add, then Add custom connector.
3. Enter a name and the remote MCP server URL, then click Continue.
4. Under Authentication, choose Sign in now, Sign in when needed or No sign in. If your server uses an API key instead of OAuth, add it under Request headers.
5. Click Add.

On Team and Enterprise, an Owner adds the connector under Organization settings > Connectors, then each member connects with their own account. In a chat, click + in the lower left, hover Connectors, and toggle the ones you want. Claude reaches custom connectors from Anthropic's cloud, not your computer, so the server must be reachable from the internet.

Claude Desktop with a local server. In Claude Desktop, open the Claude menu > Settings > Developer > Edit Config. That opens claude_desktop_config.json (on macOS, in ~/Library/Application Support/Claude/). Add your server, then fully quit and restart, per the MCP local server guide:

{
  "mcpServers": {
    "my-shopify-analytics": {
      "command": "node",
      "args": ["/Users/you/shopify-mcp/server.js"],
      "env": {
        "SHOPIFY_STORE": "your-store.myshopify.com",
        "SHOPIFY_CLIENT_ID": "your-client-id",
        "SHOPIFY_CLIENT_SECRET": "your-client-secret"
      }
    }
  }
}

Local servers only work in Claude Desktop, not on claude.ai.

Why order data can't tell you where you're losing sales

Orders record the people who bought. Each order carries a customer journey summary with the buyer's first and last visit and days to conversion, but the shoppers you lost never created an order, so order data has nothing on them. "Where am I losing sales?" is a comparison between people who converted and people who didn't, and that needs session data:

• Landing page: which entry pages get traffic but few carts.
• Traffic source and campaign: referrer, channel and UTM parameters.
• Device and funnel step: where mobile or desktop sessions stop, from cart to checkout to purchase.
• New versus returning: first-time visitors often need a different page than repeat customers.
• Multi-visit paths: a shopper who clicks a Meta ad Monday and buys from a Google search Thursday. A source that links those visits into one journey credits both.

Shopify does record sessions: the ShopifyQL sessions dataset has landing page, referrer, UTM, device and funnel-step fields. Since late September 2026, Shopify filters identified bot sessions out of session reports by default; in ShopifyQL, add WHERE human_or_bot_session = 'human' to match. So options 1, 4 and 5 can answer lost-sales questions, option 3 can't, and option 2 depends on whether it returns session data when you test it. For what a normal rate looks like, see our Shopify conversion rate benchmarks.

Questions to ask once you're connected

1. Which landing pages had the most sessions but the lowest conversion rate in the last 30 days?
2. Where do sessions drop off most: cart, checkout, or purchase?
3. How does conversion compare between mobile and desktop, by landing page?
4. Which sources and campaigns bring sessions that add to cart but don't buy?
5. How do first-time visitors from paid social convert compared with returning visitors?
6. Which products get the most views but the fewest add-to-carts?
7. How many visits does a typical buyer make before purchasing, and which source starts those journeys?
8. Which Meta campaigns convert below the store average, and which pages do they land on?
9. How do returning customers behave on the homepage compared with first-time visitors?
10. Which three segments would recover the most revenue if they converted at the store average?

The answers usually point at a page mismatch: cold traffic landing on a product page written for people who already know you, a Meta ad whose promise the page doesn't continue, or returning customers seeing a homepage built for strangers. A dedicated landing page is the usual first fix.

Security: keep it read-only and revocable

• Prefer read-only. Decline write access in the Shopify approval screen, and request only read_ scopes in a custom app.
• Use a limited staff account. Shopify limits a connected tool to the permissions of the account it acts through.
• Never paste tokens or client secrets into a chat. Keep them in your server's environment, or in Claude's Request headers field for a custom connector.
• Connect only servers you trust. Anthropic warns about prompt injection, and OpenAI warns that an untrusted server may steal information or trick ChatGPT into using tools in unintended ways.
• Ask for aggregates, not customer lists. You rarely need names or emails to find a leak.
• Know how to revoke. Uninstall the integration from your Shopify admin, remove the connector under Customize > Connectors in Claude, and for a custom app, rotate the client secret in the Dev Dashboard (Settings > Credentials > Rotate, then Revoke the old one once tokens are replaced).

Which option to pick

For sales, order and product questions, use Shopify's official plugin or connector. For a one-off funnel review, export the session reports. For "which pages and segments are losing sales," you need a live session source: build on read_reports if you have a developer, or connect an analytics app's MCP server.

That last question is the one Ariven was built around: ask which segments, sources and pages are losing sales, and follow up like you would with an analyst. When the answer is "first-time visitors from this ad need a different page," you build that version in the theme editor, and the AI can set up the flow in Ariven that routes that segment to it. Ariven doesn't build or pick the page; you do.

Sources

• Shopify Help Center: Connecting AI tools, Shopify plugin for ChatGPT, Shopify connector for Claude, Authorizing AI tool access, Shopify CLI connector, Sidekick, Exporting reports, Behavior reports, Acquisition reports, Session measurement update, Custom apps
• Shopify.dev: Access scopes, shopifyqlQuery, ShopifyQL sessions dataset, CustomerJourneySummary, Create apps using the Dev Dashboard, Application-only access tokens, Client secrets, Protected customer data, Shopify AI Toolkit, Migrate from Storefront MCP
• OpenAI: Add custom MCP server, Connect and test your plugin, Connecting custom MCP servers in ChatGPT
• Anthropic: Custom connectors using remote MCP, Use connectors, Introducing the Model Context Protocol
• Model Context Protocol: What is MCP, Connect to local MCP servers
• Zapier: Zapier MCP, Shopify MCP, Zapier MCP docs, Shopify integration. Make: Make MCP server