Built for Humans and AI Agents: Our First Look at Vercel Shop
With Vercel Shop, there is now an official open source template for Shopify built on Next.js, which Vercel positions as the template for the new Hydrogen. Over the last few days we took a closer look, set up a project and played around with it a bit.

Islam Mechtijev
Director Web & DevOps, Co-Founder @ Digit-One
If you have ever built a headless shop, you know the pattern: before the first real feature goes live, you rebuild product pages, category pages, search, cart, sitemap and caching for what feels like the fifth time. Vercel Shop ships exactly that foundation. What kept us busier while trying it out, though: the template is designed from the ground up so that AI agents can help build it and shop in it.
At Digit-One, we work with Shopify, Next.js and Sanity. every day. Here are our first impressions: what's inside, what we liked, where it falls short and who it's worth it for.
What Vercel Shop actually is
Vercel Shop is a headless storefront for Shopify, built as a regular Next.js app. The division of labor is clear: Shopify manages catalog, checkout and payments, Vercel Shop provides the frontend in front of it. The code is open source and available on GitHub, and a live demo is running at template.vercel.shop.
Out of the box, the template includes:
- Product, collection and search pages powered by the Shopify Storefront API
- Blog, pages and policies straight from Shopify
- A server-side cart with handoff to Shopify Checkout
- Navigation and footer, optionally with menus managed in Shopify Admin
- SEO, sitemaps and structured data for products and collections
- Machine-readable content for LLMs
- Webhook-based revalidation, so changes in Admin show up in the shop right away
- Optional customer accounts and an AI shopping assistant
The context matters: Vercel and Shopify have rebuilt Hydrogen together. The new Hydrogen remains open source but is no longer tied to a specific framework or runtime. It works with Next.js, Nuxt, Svelte or other frameworks. Vercel Shop uses the framework-agnostic Hydrogen SDK for API clients and cart, while Next.js handles routing, rendering and caching. It was officially presented at Vercel's Ship 26 conference on October 15, 2026 in San Francisco, together with Shopify CEO Tobi Lütke.
Getting started works the way developers expect from a modern stack today: pull the template via create-next-app, install the Headless sales channel in Shopify, put the Storefront Access Token and Shop ID into .env.local, run pnpm dev, done. Shortly after, our products were running locally in the browser, without boilerplate and without a configuration marathon. If you are starting from scratch, a Deploy button creates a new Shopify store and a Vercel project in one go. Requirements are Node.js 24 and pnpm 12.


Agent-native: AI builds it, AI shops in it
"Agent-native" sounds like marketing at first. After a few hours with the template, we can say it is meant quite literally. Vercel Shop considers AI agents on two sides.
Coding agents that extend the shop
Every project includes an AGENTS.md with the template's conventions and a collection of skills in .agents/skills/. Skills are small, task-specific instructions with checks that coding agents like Claude Code, Cursor or Codex work through. Claude Code picked them up right away for us, without any extra configuration. On top of that, the Shopify AI Toolkit gives the agent current API documentation and schema validation. So the agent doesn't guess which GraphQL fields exist, it checks them against Shopify.
Some of the included skills (https://vercel.shop/docs/getting-started/extending-with-agents):
build-shop: Build routes, data fetching and rendering following the template's patternsupdate-shop: Compare your shop with the current template and adopt updates selectivelyshopify-graphql-reference: Integrate validated Storefront and Customer Account queries cleanlyenable-shopify-markets: Multiple regions with their own currency and commerce contextenable-i18n: Translated UI with a locale prefix in the URLenable-shopify-menus: Manage navigation and footer from Shopify Adminenable-analytics: Vercel Analytics, Speed Insights and Google Tag Manager
What we like about it: an extension like "size chart from a metafield below the product description" follows the same pattern every time, no matter who on the team or which agent implements it. That is exactly what we mean when we talk about AI with guardrails.
Shopping agents that buy in the shop
The second side concerns shopping agents, such as ChatGPT, Perplexity and the like, that search for products on behalf of people. Vercel Shop makes the shop readable for them without changing the pages for human visitors. We checked this out directly in the demo:
- Markdown instead of HTML. Home, products, collections and search return structured Markdown when a client asks for it or when you append
.mdto the URL. Try it yourself: just append.mdto a product URL in the demo. These versions are set tonoindexand point to the HTML page via canonical, so Google keeps indexing the right page. - llms.txt. A compact overview of the shop for AI agents, with links to policies, collections and the sitemap.
- Universal Commerce Protocol (UCP). Shopify provides catalog search, live availability per variant, cart and checkout via UCP, and Vercel Shop forwards these endpoints on your own domain.
- WebMCP. Experimental and off by default: search and cart as structured tools for agents in the browser.

There is also the optional Shop Agent, a built-in chat assistant based on Vercel's agent framework Eve. It finds products, answers questions about shipping or returns and updates the cart. It cannot place orders or pay, the final click stays with the human. Vercel's own merch store shows what this looks like live.
For us, this isn't theory: we have already implemented llms.txt for our own website, and AI visibility now comes up in almost every client conversation. A shop template shipping this out of the box is a clear signal.

Mini-Glossar
- llms.txt: An overview file that explains to AI agents what can be found on a website.
- UCP: The Universal Commerce Protocol by Shopify, through which agents can search for products, check availability, and fill a shopping cart.
- WebMCP: An experimental browser standard through which websites offer agents structured tools.
- Skills: Guides with checks that coding agents work through step by step.
This is what the shop looks like for AI agents
- Open demo: template.vercel.shop
- Click on a product and append
.mdto the URL - Append
/llms.txtto the domain - Test the shop agent live: Vercel Merch Shop
Fast by Default: Architecture, caching and security
Performance in Vercel Shop isn't an afterthought, it is built into the architecture. Clicking through the demo and looking at the code, these decisions stood out to us:
- Shell first. Page title, main image, price and availability of the selected variant come as early as possible. Recommendations, filters or account data load later in their own Suspense boundaries. The page doesn't wait for the slowest Shopify request.
- Caching with clear rules. Public data such as products, collections and menus runs through Next.js Cache Components and cache tags. Cart, session and customer data never end up in a shared cache.
- Webhook revalidation. When someone changes a product in Shopify Admin, a signed webhook invalidates exactly the affected content. Without webhooks, product data can stay cached for up to roughly 30 days according to the docs, so setting them up is a must, not a nice to have.
- Clean variant URLs. Color and size appear as readable parameters in the URL, such as
?Color=Blue&Size=XS. Old Liquid links with?variant=are redirected automatically, and Shopify paths like cart permalinks or/cart.jskeep working. For migrations from existing themes, that's a real plus.
When it comes to security, the setup plays to its strengths: checkout and payment data stay entirely with Shopify, and card details never touch the storefront. Customer sessions are encrypted and stored in HttpOnly cookies, API secrets stay on the server. On Vercel, firewall and DDoS protection come on top, optionally also the invisible bot detection BotID.
One thing to keep in mind: these are architectural decisions, not benchmarks. How fast a specific shop ends up being depends, as always, on images, third-party scripts and your own code.
Required After Setup
Set up webhooks. Without them, product data can remain cached for up to about 30 days.
Register these topics:
products/create,update,deletecollections/create,update,deletemetaobjects/create,update,delete
Where it fits: why headless, and what to watch out for
For many shops, headless was long a case of "really interesting, but too much effort". Vercel Shop significantly weakens exactly that argument. Anyone who wants more than a theme can offer, such as their own brand worlds, lots of content, custom integrations or top performance, now gets a solid foundation instead of an empty repo. A well-built theme is still a valid choice for many merchants, the difference lies mainly in the room to grow:
Shopify Theme (Liquid)
- Hosted by Shopify, no operations of your own
- Content in the theme editor in Shopify Admin
- Design within the theme logic
- Integrations via Shopify apps
- AI readability depends on theme and apps
- Quick start, but limits when scaling up
Vercel Shop (Headless)
- Your own Next.js project hosted on Vercel, including edge, firewall and scaling
- Commerce in Shopify, content in a CMS like Sanity, cleanly separated
- Full design freedom, every page and every component lives in your own code
- Shopify apps plus your own routes, search, analytics and any API
- Markdown, llms.txt and UCP out of the box, ready for agentic commerce
- More initial effort, which the template reduces significantly
Good to know before the project starts: a template is not a finished shop. While trying it out and working through the docs, we noticed a few points you should plan for. None of them is a showstopper, but all of them belong in project planning:
- A lot is switched off on purpose. Customer accounts, Shop Agent, analytics, bot protection and WebMCP are disabled by default. That's actually nice, because only what you really need is running, but each feature needs its own configuration.
- Webhooks don't cover everything. Products, collections and metaobjects are updated instantly. Pages, blog articles, policies, menus and inventory are not, so you need your own strategy here. With Sanity as the CMS, editorial content is covered anyway.
- One language, one market. The default is one deployment with one language. i18n and Shopify Markets are added via skills, and for the DACH region that's practically always part of the project.
- The Shop Agent needs guardrails. The chat is publicly accessible, and model costs have no built-in cap. Conversation content goes to the model provider, and conversations are not tied to a customer by default. For European shops, that means rate limits, spending limits and a proper GDPR review before the assistant goes live.
- Customer accounts need HTTPS and fixed URLs. Shopify doesn't accept wildcards for callback URLs, so every production and preview domain has to be registered individually. Locally, this only works through a public HTTPS tunnel.
- The Headless sales channel is mandatory. Products published only to the Online Store don't show up in the storefront. That's the classic pitfall during the first setup: if the home page stays empty, this is almost always why.
Disabled by default
- Customer Accounts
- Shop Agent
- Analytics
- Bot Protection
- WebMCP
How we plan to work with it at Digit-One
Day to day, we rely on Shopify for e-commerce and on Next.js with Sanity for web projects. When a shop needs more than a theme can offer, we combine all three into our full stack for headless commerce. Vercel Shop fits right in, because the template is explicitly designed for you to add your own CMS, search and analytics alongside it. After our first tests, here's our plan:
- Shopify remains the single source of truth for commerce. Prices, availability, cart and checkout come from Shopify, just as the template intends. We don't touch that.
- Sanity handles the storytelling. Landing pages, brand worlds, lookbooks and editorial content are managed by marketing in Sanity, linked to products from Shopify. Editors get a comfortable editing experience without product data being maintained twice.
- The template is the starting point, not the result. We take over the foundation and build what sets the shop apart on top of it: design, content modules, integrations. With the
update-shopskill, improvements from the original stay easy to pull in. - Custom skills for client projects. The docs themselves recommend capturing recurring conventions as project skills. That's exactly how we work: every shop gets its own rules for components, data fetching and tests, so AI-assisted development stays consistent.
- AI visibility from the start. We configure Markdown responses, llms.txt and structured data from day one instead of retrofitting them later. Anyone building a shop today is also building it for shopping agents.
Not every shop needs headless. For many merchants, a well-built Shopify theme is still the more economical choice, and we say so. Vercel Shop gets exciting for brands with lots of content, multiple markets, custom integrations or the ambition to be early in agentic commerce. And if you already have another system in place, you don't have to throw everything away: we look at what's there and optimize where it pays off.
Our Full Stack for Headless Commerce
- Shopify for Catalog, Cart, and Checkout
- Next.js with Vercel Shop as the Frontend
- Sanity for Content and Storytelling
- Vercel for Hosting and Deployment
Conclusion
After our first experiments, Vercel Shop is the most convincing argument so far for building headless commerce on Shopify with Next.js. The template takes the grunt work off your hands, sets sensible defaults for caching and security and makes the shop readable for AI agents from day one. The fact that Vercel and Shopify are driving this together gives it a perspective that community starter kits rarely have.
At the same time, it's a template, not a finished shop. i18n, Markets, content strategy, privacy for the assistant and the actual design remain project work. That's exactly where we come in, and we'll keep following the topic.
Thinking about taking your Shopify shop headless or getting it ready for AI shopping? Get in touch, we'll look at your setup together and tell you honestly whether the step is worth it.
