Nic Dunn, CEO, Charle Agency
The biggest Shopify change of 2026, in one paragraph
Three releases sit behind the headline. In December 2025, Shopify shipped its Dev MCP Server inside the Winter '26 Edition, giving AI coding agents structured access to the platform. On 24 March 2026, Agentic Storefronts went live by default for eligible US merchants, making 5.6 million stores automatically discoverable inside ChatGPT, Microsoft Copilot, Google AI Mode and the Gemini app. Then on 9 April 2026, Shopify open-sourced the AI Toolkit, the consumer-facing wrapper for the Dev MCP. Taken together, these moves opened Shopify's commerce platform to AI on both ends. Shoppers can now buy through Claude and ChatGPT. Merchants and developers can build and run their stores by talking to an agent.
What is Shopify MCP, exactly?
The Model Context Protocol, or MCP, is an open standard developed by Anthropic for connecting AI models to external tools and data sources. Anthropic released it as open source in late 2024, and it has since been adopted across every major AI platform. Shopify built its commerce-side AI infrastructure on top of MCP rather than inventing a proprietary protocol. That choice matters, because it means the same store can connect to Claude, ChatGPT, Cursor, Gemini, or any other MCP-compatible client without a custom integration for each.
When people talk about Shopify MCP, they usually mean one of three different things, and the confusion is widespread.
The first is the Storefront MCP server. Every Shopify store now has its own endpoint at /api/mcp. This is what AI agents call when a shopper asks about your products inside a chat interface. It handles search, cart operations, policy questions and checkout.
The second is the Dev MCP server, which Shopify ships through the AI Toolkit. This connects developer-facing AI tools like Claude Code and Cursor to Shopify's documentation, GraphQL schemas and code validation. It is how engineering teams build apps and themes against Shopify with an AI agent in the loop.
The third is community-built Admin MCP servers. These are open source projects, like GeLi2001's shopify-mcp, that connect Claude or other clients to the Shopify Admin GraphQL API for direct CRUD operations on products, orders, customers and inventory. They are not made by Shopify, and they require a Shopify access token to operate.
All three use MCP. They serve very different jobs. Get this distinction right and the rest of the picture clicks into place.
The two layers of Shopify's AI Toolkit (and why both matter)
Most coverage of Shopify and AI focuses on one half of the picture. Either it talks about agentic shopping, where consumers buy through ChatGPT, or it talks about the developer toolkit, where engineers write code with Claude. Both are real. Both ship through the same protocol. And both demand attention from a Shopify Plus brand at the same time.
We think of these as two parallel commerce shifts.
The first is the consumer-facing layer. Storefront MCP is how AI assistants now interact with your store on behalf of a shopper. The shopper does not visit your site. They have a conversation. Your product catalogue, policies, inventory, pricing and shipping information all need to be legible to an AI agent that may never render a single pixel of your theme. This is a discoverability and conversion shift. It changes how brands earn visibility in the new search.
The second is the operator-facing layer. The AI Toolkit gives developers and operators a structured way to build, validate and run Shopify code inside their AI tools. It accelerates how fast a team can ship store improvements, theme changes, app integrations and bulk operations. This is a velocity shift. It changes how quickly brands can iterate.
A brand that only optimises for the first will be discoverable but slow. A brand that only optimises for the second will move quickly but stay invisible to the new buyers. The compounding advantage sits with the brands that take both seriously, and that is where the strategic work lives.
The Storefront MCP: how shoppers find and buy through AI
Every Shopify store has a live MCP endpoint at https://your-store.myshopify.com/api/mcp. No setup. No app to install. This has been the case across the platform since the Summer '25 Editions release.
Through that endpoint, AI agents can do four things on behalf of a shopper:
- Search your product catalogue with natural language queries like "running shoes under £100 in waterproof leather"
- Read and update the shopping cart
- Answer questions about your shipping, returns and policy pages
- Initiate checkout
The Customer Accounts MCP server adds a second layer for logged-in shoppers. It handles order tracking, returns and account information, with authentication routed through Shopify's customer account flows.
The numbers behind this matter. In its Q3 2025 earnings call, Shopify reported AI traffic up 7x and AI-attributed orders up 11x since January of that year. A Shopify survey found that 64% of shoppers said they were likely to use AI to some extent when making purchases. McKinsey, in its October 2025 research, projected $3 trillion to $5 trillion in global agentic commerce orchestration by 2030, with the US B2C retail market alone reaching up to $1 trillion.
Then on 24 March 2026, Shopify activated Agentic Storefronts by default for eligible merchants. 5.6 million Shopify stores became automatically discoverable inside ChatGPT, Microsoft Copilot, Google AI Mode and the Gemini app. As Shopify CEO Tobi Lütke put it: "We're making every Shopify store agent-ready by default." For US merchants whose products are in the Shopify Catalog, ChatGPT shoppers can already find them.
This is now the new front of the funnel. A shopper opens Claude or ChatGPT, asks a question, and a list of candidate products comes back. The brands that show up are the ones whose product titles, descriptions, metafields, FAQs and policy pages are clean and legible to a language model. The brands that do not show up are not failing because their SEO is broken. They are failing because their data structure was written for human eyes, not for a machine that has to choose one of a thousand options to recommend.
Charle's view: Most Plus brands are about to discover the gap between what they tell shoppers about their products and what an AI agent can actually find and recommend. The brands closing that gap in 2026 will own the new search results. The ones who wait will spend 2027 trying to catch up.
The Dev MCP and AI Toolkit: how teams build and run Shopify stores faster
The Shopify AI Toolkit, open-sourced on 9 April 2026, is the developer-side of the same MCP infrastructure. It is a free, MIT-licensed plugin that connects AI coding agents like Claude Code, Cursor, VS Code, Gemini CLI and OpenAI Codex directly to Shopify's documentation, GraphQL schemas, and live store operations through the Shopify CLI.
Inside the toolkit are seven distinct tools that an agent can call during a development workflow:
learn_shopify_apigives the agent a structured overview of the platform's API surfacesearch_docs_chunksandfetch_full_docslet the agent pull relevant documentation in real time, rather than relying on whatever was in its training dataintrospect_graphql_schemaqueries Shopify's GraphQL schema directly, so generated code is aligned with the current API rather than an outdated snapshotvalidate_graphql_codeblocks,validate_component_codeblocksandvalidate_theme_codeblockscheck generated GraphQL queries, Hydrogen components, and Liquid theme sections against Shopify's actual schemas before the code runs
There is also a store execute capability layered on top, which uses the Shopify CLI to perform real operations on a connected store. That includes creating and updating products, managing metafields, modifying theme files, and running bulk operations. The CLI side requires Shopify CLI authentication and the appropriate Partner permissions, so it is not unauthenticated like the documentation tools.
The point is not that AI replaces a Shopify developer. It is that the developer's loop is collapsed. In a typical workflow without the toolkit, a developer building a metafield-powered size guide will switch between API docs, schema references, the Partner dashboard, an editor and a test store. With the toolkit, the agent does the lookups, the schema check and the validation in one conversation, then writes the code that the developer reviews and ships.
A few important caveats. The official AI Toolkit is built for development, not full conversational store management. You cannot ask it to "update all my product prices by 10%" through its core tools. That kind of conversational store management runs through community-built Admin MCP servers, like GeLi2001's shopify-mcp, which connect Claude or other clients to the Shopify Admin GraphQL API. These require a Shopify access token, careful scope management, and they are not maintained by Shopify, so they can break when Shopify updates the API. They are useful, but they sit in a different risk category to the official toolkit.
Charle's view: The toolkit is real productivity, but it is not a strategy. Speed compounds when it is pointed at the right priorities, and that is the conversation we have with Plus brand teams: not "should we use this", but "given everything else you are running this quarter, where does this earn its keep first?"
How to connect your Shopify store to Claude (step-by-step)
There are two routes, depending on your setup.
The recommended route is the Shopify-published Claude Code plugin. From a terminal, with Claude Code installed, run:
/plugin marketplace add Shopify/shopify-ai-toolkit
/plugin install shopify-plugin@shopify-pluginRestart Claude Code. You now have access to all seven tools in the AI Toolkit. The plugin updates itself automatically as Shopify ships new capabilities, so you do not need to manage versions.
The manual route uses the Dev MCP server directly. From your terminal:
claude mcp add --transport stdio shopify-dev-mcp -- npx -y @shopify/dev-mcp@latestThis adds the MCP server but does not give you the auto-updating plugin layer. You manage updates yourself. Restart Claude Code after the add command.
Either route requires Node.js 18 or higher. Neither requires authentication for the documentation or schema tools. They run locally and connect to publicly available Shopify developer resources.
If you want store execute capabilities (creating products, managing metafields, modifying themes against a real store), install the Shopify CLI separately with npm install -g @shopify/cli, then authenticate with shopify auth login. Make sure your Partner account has the appropriate access to the target store, and never connect to a production store without a review process in place.
A practical first test once it is running: ask Claude something like, "Use the Shopify AI Toolkit to explain how metafield definitions work and show me the GraphQL mutation for creating one." If the response cites real documentation pages and includes a validated mutation, the connection is live.
For Shopify Plus brands with development partners, the more interesting setup conversation is who has access to which store and how that access is governed. The toolkit's productivity gain is real, but a centralised Partner-account approach with named developers, audit trails and staging-first workflows matters more on Plus than on a single-merchant store. That is the kind of governance question we work through with brands as part of an engagement.
How to connect your Shopify store to ChatGPT and agentic storefronts
The consumer-side connection works differently to the developer-side, and most merchants miss this.
There is no plugin to install. The Storefront MCP endpoint at /api/mcp is already live on every Shopify store. What you control is whether your store is set up to be found and recommended by AI shopping agents.
For US merchants, Agentic Storefronts went live by default on 24 March 2026. If your store is eligible, sells to US shoppers, and has products in the Shopify Catalog, ChatGPT can already surface your products in shopping queries. The same is true, with rolling rollouts, for Microsoft Copilot, Google AI Mode and Gemini. UK and EU merchants are not yet on the same default-on schedule, though Shopify has signalled that international rollout is coming. We are watching this carefully for our UK Plus clients.
To test what an AI agent currently sees when it queries your store, you can hit your own MCP endpoint directly. Send a JSON-RPC POST request to https://your-store.myshopify.com/api/mcp with a tools/call method and a search_shop_catalog argument. The response shows the products, prices and metadata that an AI agent would surface. If the data looks thin, inconsistent, or full of promotional language that does not describe the product, that is exactly what ChatGPT shoppers are seeing.
The control points that move the needle here are the ones we work on across our Plus engagements: structured product data, clean metafields, policy pages written in plain English with answerable questions, schema markup for trust signals, and brand-defining content that an AI agent can cite as a reason to recommend your store over a competitor. None of this is new SEO work in principle. It is SEO done with AI legibility as the primary success metric.
What this means for Shopify Plus brands in 2026
The strategic conversation we are having with Plus clients right now revolves around three shifts.
First, catalogue hygiene becomes AI discoverability hygiene. The basic cleanliness work that brands have always known they should do (tight product titles, accurate descriptions, structured metafields, clean variant data, consistent inventory feeds) now has direct revenue consequences. An AI agent recommending products parses exactly that data. Shopify Plus brands typically have larger, messier catalogues built up over years of campaigns and migrations. The brands that audit and rebuild this quietly in 2026 are the ones AI assistants will choose in 2027.
Second, development velocity becomes a competitive moat. The AI Toolkit is real productivity. A team that integrates it into its workflow can ship store improvements, A/B tests, integrations and bulk operations measurably faster than a team that does not. For a Shopify Plus brand running campaigns, peak periods, international rollouts, and B2B alongside DTC, that velocity compounds fast. Twelve months from now, the brands that started using the toolkit early will have shipped two or three times as many tested experiments as the brands that waited.
Third, brand-defining content moves up the priority list. AI agents do not just match keywords. They make recommendations based on whatever signals they can find about why your brand is worth choosing. Editorial content, founder POV, certifications, third-party reviews, unique product context, the specifics of how you ship or how you handle returns. All of it influences whether your store gets surfaced. The thin, generic copy that has been quietly tolerated on many Plus catalogues becomes a real liability when the new front of the funnel is a conversation.
In our experience working with Shopify Plus brands across DTC, B2B and international, the teams that pull ahead in this kind of platform shift are not the ones who attempt to do everything internally. They are the teams that pair internal commercial knowledge with a partner who has seen the platform shift across multiple brands and can sequence the work properly. That is the role we play.
Limitations, risks and what's coming next
A few honest caveats before any Shopify Plus brand jumps in.
There is no mobile MCP support. The official AI Toolkit runs in desktop AI clients (Claude Code, Claude Desktop, Cursor, VS Code). For mobile-based AI store management, Sidekick is the option Shopify offers, and Sidekick itself is built on Claude Sonnet 4.5 through Google Cloud's Vertex AI, according to Andrew McNamara, Shopify's Director of Applied AI.
Rate limits are real. Shopify's GraphQL Admin API allows 1,000 cost points per minute on standard plans, with higher limits on Shopify Plus. Heavy automation workflows (bulk operations, schema introspection on large catalogues, repeated validation calls) can hit these limits faster than expected, and your AI agent will not always handle the back-off cleanly.
Community Admin MCP servers carry maintenance risk. They depend on individual maintainers tracking Shopify API changes. When the API moves and the maintainer is on holiday, things break in production. We treat them as useful infrastructure for staging environments and internal tools, not as core production dependencies.
Security and access discipline matter more, not less. Access tokens should never be hardcoded or committed to version control. Delete operations through MCP are irreversible. Shopify's VP of Engineering Farhan Thawar has written about "comprehension debt" when teams over-rely on AI: if the developer cannot debug the code the agent generated, the team carries hidden risk. None of this rules out the toolkit. It just means the governance has to keep up.
What is coming next. Sidekick's write capabilities are widening, MCP UI components are landing, and the international rollout of Agentic Storefronts is only a matter of time. Twelve months from now, the picture will look meaningfully different again.
Charle's view: The biggest risk for Plus brands here is not technical, it is governance. Who has access to which store, who reviews what the agent ships, what is allowed in production versus staging. These are the decisions that prevent bad outcomes, and the toolkit makes them more important, not less.
Where to start this week
If you are running or leading a Shopify Plus brand, three concrete actions are worth booking time for in the next fortnight.
Test your store's MCP endpoint. Hit https://your-store.myshopify.com/api/mcp with a search query and look at what comes back. Show the response to your team. The first time most operators see this, the gap between "what we tell shoppers" and "what an AI agent actually sees" becomes obvious.
Audit your product data and policy pages for AI legibility. Pick your top 50 SKUs by revenue and your most-trafficked policy pages. Check titles, descriptions, metafields, FAQ wording, returns and shipping clarity. If a sentence is ambiguous to a human, it is unanswerable to a machine. Fix those first.
Decide who in your team owns the AI surface. This is the part that gets skipped. Catalogue legibility, agentic discoverability, and AI-assisted development are not a single team's responsibility today inside most brands. They cross product, marketing, content and engineering. The brands that win in 2026 are the ones who name an owner now.
This is where our work with Plus brands is concentrated right now: the audit, the prioritisation, and turning the platform shift into a roadmap a brand can actually execute against alongside everything else it is doing. If that would be useful to compare notes on, our Shopify Plus team is happy to talk.
