Shopify MCP: Shopify's own servers, and store revenue next to ad spend

A Shopify MCP server is any server Shopify runs that speaks the Model Context Protocol, and Shopify runs four. The one a merchant means is the Shopify connector: a remote server at setup.shopify.com/mcp, made by Shopify, listed as Anthropic verified in Claude's directory since April 2026 with 25 tools, and available in ChatGPT as the Shopify plugin. Shopify's Help Center says it can build a store from a storefront preview, manage a store you own, and source products from Shopify Collective suppliers. Shopify also documents a connector for Perplexity, which manages an existing store only, and one for Muse.

The other three are for developers. The Dev MCP server, started as a package in March 2025 and at version 1.16.0 on 25 September 2026, runs on your computer with npx, needs no authentication, and lets a coding agent search Shopify's documentation and API schemas; it never touches a store. The Storefront MCP at {shop}/api/mcp let agents search a catalog and manage carts; its catalog and cart tools were removed in favour of the Universal Commerce Protocol at {shop}/api/ucp/mcp, announced on 24 June 2026 with the old tools kept until 31 August 2026. The Customer Accounts MCP at {shop}/customer/api/mcp answers a signed-in customer's own questions, order status and account details, for an app that has been granted protected customer data.

So the question behind the search is really two questions. If you want an assistant to change the store, the connector is the answer and this page describes its limits. If you want an assistant to read the store next to your ads, that is what the Adrails MCP server does, and the second half of this page describes those tools.

• Shopify connector: merchants, in Claude, ChatGPT, Perplexity and Muse
• Dev MCP: developers, local, documentation and schemas only
• Storefront MCP, now UCP: developers building a shopping agent
• Customer Accounts MCP: apps acting for a signed-in customer

The connector's 25 tools are named on its directory page: search_products, get-shop-info, get-product, list-orders, get-order, list-customers, get-inventory-levels, run-analytics-query, search_collections and get-collection on the reading side; create-product, update-product, set-inventory, create-discount, add-to-collection, bulk-update-product-status, create-collection, update-collection, upload-image and switch-shop on the writing side; and graphql_schema, graphql_query, graphql_mutation and validate_graphql_codeblocks for anything the Admin API allows. Shopify's Help Center spells the reads out: the 50 most recent orders with customer, totals and payment and fulfilment status, an order's line items and tracking, inventory by location, a customer's order count and total spend, and analytics on sales, orders, products, customers and fulfilment.

The writes are concrete too: a product with title, description, price, variants and options, products set to active, draft or archived in batches of up to 50, manual and smart collections, a quantity at one location with a reason, and a percentage-off discount code. Shopify is as clear about what the connector cannot do: issue refunds, cancel or capture orders, mark orders as paid, process returns, create or adjust gift cards, edit or publish the live theme, or change the plan, payments, taxes or domains. It also warns that a connected tool changes items one at a time, so a large batch can stop partway.

Access follows two rules. On the install page you allow a level of data access, some of it read-only and some with write access, and the tool is then limited by the permissions of the staff account it acts through: a user who can read products but not edit them cannot give the tool the ability to edit them. Shopify asks you to read the AI tool's terms before connecting, reminds you that data shared with the tool is no longer processed by Shopify, and holds you responsible for a change the tool makes. Access is revoked by uninstalling the integration from the Shopify admin. There is no fee for the connector; standard Shopify plan pricing applies.

Adrails connects to Shopify with four read scopes and nothing else: read_orders for orders, refunds and their status; read_customers for customers and their order history; read_products for products, prices and inventory; read_reports for sessions, conversion rate and the checkout funnel. You paste the address of your Shopify admin under Integrations and allow the access in Shopify. Adrails never writes to the store, and uninstalling the Adrails app from Shopify ends the connection.

Orders arrive in real time through Shopify webhooks and are reconciled every 15 minutes. Customers and products are read every hour, store analytics every 6 hours. Customer email addresses are turned into a one-way fingerprint before they are stored, so Adrails keeps no readable address, and revenue stays in each store's own currency, never converted.

In the app this feeds three places. Analysis shows sessions, net sales, orders, conversion rate, average order and the store funnel next to ad spend, with Shopify's figures shown next to the platforms' figures rather than merged into them. A campaign's Tracking tab shows the customers each ad brought and their lifetime spend. Automations ship with the Best sellers and Ads against store templates, which post on a schedule to Slack, Discord or email.

Once Adrails is added as a custom connector, your client sees 63 tools; the commerce ones read the Shopify connection. store_revenue returns revenue over a period with gross, refunded and net kept apart and the order count, in each store's own currency. store_orders lists orders newest first with totals, financial status, fulfilment status and line count, at most 50 per call. store_customers returns name, email domain, country, creation date, order count and imported lifetime spend, with full addresses never stored. store_products returns title, price range, inventory, status, type and vendor, filtered on the title.

metrics_daily gives the daily series the store measures, sessions, conversion rate, sales and average order value, with a breakdown by product, traffic source or country, and called without metrics it lists everything measured so you can discover the names. Next to them sit the advertising reads: reporting_overview for spend, attributed revenue and conversion events across every connected account, performance_period for one account's figures over an exact range with a comparison range, and business_review, which checks a revenue reconciliation or a unit economics calculation from figures quoted word for word from earlier results.

The changes are proposals, not writes. campaigns_pause, campaigns_change_budget, campaigns_duplicate_to_draft, campaigns_split_test and campaigns_launch each prepare one exact change and return a link; the person who connected, as an owner or admin on an active plan, reviews it field by field in Adrails and applies it or declines it within 30 minutes. A budget moves by at most 30% per change, everything Adrails creates on a platform starts paused, and an applied change can be undone while the platform still matches what Adrails wrote. A tool call uses no agent credit, since your client's model does the reasoning.

• store_revenue, store_orders, store_customers, store_products: the Shopify reads
• metrics_daily: daily store series, broken down by product, source or country
• reporting_overview and performance_period: the ad side for the same dates
• business_review: the reconciliation checked from quoted figures
• campaigns_pause, campaigns_change_budget, campaigns_launch: proposals applied in Adrails

Connect the Shopify connector when the work is the store: a product to create, inventory to set, a discount code for the weekend, a collection to reorganize, a question about one order. It is the only one of the two that writes to Shopify, and it reads everything the Admin API exposes through GraphQL. It knows nothing about your ads, and it lets the assistant change the catalog under your staff permissions, so Shopify's own advice applies: read the tool's terms, watch the changes it makes, and block the write tools in your client when you only want to read.

Connect Adrails when the question joins the store to the ads: what the store sold against what the ads spent, which customers an ad brought and what they are worth, whether Meta's attributed revenue is above the orders Shopify recorded, which campaign to pause or scale on that basis. Adrails reads Shopify read-only, keeps the source on every figure, and turns the decision into a proposal with a review step rather than writing to Meta or Google Ads from the chat.

Many stores will run both in the same Claude or ChatGPT account, and the two do not overlap: one changes the store, the other reads it next to the ads and prepares the ad changes. Agencies tend to connect Adrails first, because one workspace holds several Meta and Google Ads accounts next to the store they sell for, while the Shopify connector carries a switch-shop tool for moving between stores.

For the Shopify connector, open its page in Claude's directory from a desktop browser, choose Add to Claude, sign in to your Shopify admin and allow the data access it requests; in ChatGPT, install the Shopify plugin from its plugin page and allow the same access when the Shopify admin opens. Then open a new chat, enable the connector and start with a question whose answer you know, such as yesterday's order count.

For Adrails, first connect Shopify inside Adrails: open Integrations, choose Shopify, paste your admin address and allow the four read scopes. Then add Adrails to your client. In Claude, open Customize, then Connectors, choose Add custom connector, name it Adrails and paste https://adrails.ai/api/mcp; connect it, sign in to Adrails, choose the workspace and the permissions, and allow access. In ChatGPT, turn on Developer mode, add a custom MCP server with the same URL and OAuth, and review the tools it discovers. Claude Code takes one command, claude mcp add --transport http adrails https://adrails.ai/api/mcp, then /mcp to sign in.

The consent screen offers Read workspace data and Stay connected first; select Prepare advertising changes only if you want the assistant to prepare pauses, budget moves and launches for you to apply. Your first question is the one from the table: what did the store sell yesterday, and which ads paid for it.