Start free
Shopify · MCP

Your store, addressed
as agent tools.

A chatbot waits for a shopper. An MCP server answers software. Clearly exposes your Shopify catalog, product detail, cart and checkout as typed tools any agent can call — so buying can happen without a person parked in a chat window.

Install on ShopifyRead the MCP docs ⟶
A honeycomb lattice: each cell one addressable product in the catalog, with a query sweeping across it
The shift

Storefronts were built for eyes. Agents do not have any.

Every storefront convention — the grid, the filter rail, the sticky cart — exists to help a person scan a page. An agent has no eyes to scan with. Handed a storefront, it does the only thing available: fetches HTML and guesses. That is slow, brittle, and breaks the moment you reskin your theme.

Scraping a storefront

The agent parses markup meant for a browser. It cannot tell a badge from a price, a sold-out variant from an available one, or a recommendation from the thing you searched for. Change your theme and every assumption it made is void.

Calling typed tools

The agent asks search the catalog for a linen shirt under £80 and gets structured results — variant, price, stock, image. It resolves one specific purchasable thing, then asks for a checkout link. Your theme is irrelevant.

The tools

What an agent can actually do

Six verbs, each mapping to a real operation against your store. No general-purpose 'ask the store anything' box — specific tools with specific contracts, which is what makes agent behaviour predictable.

Search the catalog

Natural-language or structured queries across products, variants, tags and price bands. Returns what is purchasable now, not what was in stock last week.

Resolve a product

Full detail for one item — variants, options, price, availability, media — so the agent commits to a specific SKU rather than a product family.

Build a checkout link

A real Shopify checkout URL for a chosen variant and quantity. The shopper lands in your normal checkout, with your payment methods and your tax rules.

Create a draft order

For quotes, wholesale and assisted sales: the agent assembles the order and hands a draft to you or the customer instead of forcing a live purchase.

Sync your products

Catalog ingestion keeps the agent's view current. What the agent describes is what a customer can put in a basket.

Escalate to a human

When a question is outside the catalog — an order, a refund, an edge-case policy — the agent hands off with context attached rather than inventing an answer.

How it runs

Three steps, then it is callable

Step
01

Connect the store

Install from the Shopify App Store and authorise. Products, variants and availability ingest on connect and stay in sync from then on.

Step
02

Point an agent at it

Your store becomes an MCP endpoint with typed tools. Any MCP-capable client — Claude, an internal agent, your own code — discovers the tools and calls them.

Step
03

Let it transact

The agent searches, resolves a variant and returns a checkout link or draft order. Payment stays in Shopify checkout; a human still confirms.

6
typed tools exposed per store
0
storefront HTML parsed
Shopify
checkout — we never touch card data
Where it fits

Best on the stores a widget cannot reach

A chat bubble needs a page to sit on and a person to click it. These do not have both.

Headless storefronts

Hydrogen or custom front ends with nowhere to mount a widget, but a real need for structured catalog access.

Assistant-led buying

The shopper is talking to an agent somewhere else entirely. Your catalog still has to be reachable from there.

Wholesale & quoting

Draft orders let an agent assemble a basket for approval rather than pushing an immediate card payment.

Questions

Shopify MCP, answered

01What is an MCP server for Shopify?+
MCP (Model Context Protocol) is how an AI agent discovers and calls tools. An MCP server for Shopify exposes your store — catalog, product detail, cart and checkout — as typed tools an agent can invoke directly, instead of scraping your storefront HTML or guessing at a REST shape. The agent asks for what it needs and gets structured data back.
02How is this different from a Shopify chatbot?+
A chatbot is a surface: a widget a human types into. An MCP server is an interface: something other software calls. The same underlying store agent can serve both, but they solve different problems — a chatbot needs a shopper present, while MCP lets an agent shop, compare and build a cart with nobody watching.
03Can an agent actually complete a purchase?+
It can get a shopper to the checkout. The agent searches your catalog, resolves a specific variant, and returns a real Shopify checkout link or a draft order. Payment itself stays in Shopify checkout, where it belongs — we never handle card details, and a human still confirms the transaction.
04What does the agent know about my catalog?+
What you ingest. Products, variants, prices, availability and descriptions are pulled from your store and kept in sync, so answers reflect what is actually purchasable right now rather than a stale snapshot. Nothing outside your store data is invented — when the catalog cannot answer, the agent says so and can escalate.
05Does it work with a headless storefront?+
Yes, and this is where it fits best. A headless store has no widget to drop a chatbot into, but it does have an agent that needs structured access to the catalog. MCP is transport-agnostic — the storefront can be Hydrogen, a custom React app, or no storefront at all.
06What happens when the agent cannot answer?+
It escalates rather than improvising. Order-specific, account-specific and policy-edge questions hand off to a human with the conversation attached, so the customer does not repeat themselves. An agent that confidently invents a returns policy is worse than no agent.

Make your catalog callable.

Install the agent, connect the store, and your products become tools an agent can actually buy from.

Install on ShopifyAgent vs chatbot