BigCommerce MCP: What Exists Today and What to Use Instead

Yes, BigCommerce has MCP support, but "BigCommerce MCP" covers three different things, and only some of them are available to you today. A shopper-facing Storefront MCP server is live, a documentation server is live, and the server meant for running your store (catalog, orders, pricing) is in closed beta. This article explains which is which, how to connect each one, and what to use when the one you need is not available.

MCP (Model Context Protocol) is an open protocol that lets AI assistants such as Claude connect to external tools and data through a standard interface. If you are new to it, start with What Is MCP? A Plain-English Guide for E-commerce Merchants.

Status notes in this article reflect BigCommerce's public documentation as of late October 2026. BigCommerce is changing this area quickly, so confirm the current state in your control panel and the official docs before you plan around it.

What BigCommerce MCP products exist today?

Three official products exist. They do different jobs and are easy to confuse.

Product	Status	What it does	Touches store data?
Storefront MCP (the docs call it "BigCommerce MCP Server")	Live, one endpoint per storefront	Shopper-facing tools: product search, cart, checkout handoff	Reads the public catalog and builds carts
Store Management MCP	Closed beta	Merchant work: catalog, pricing, inventory, orders, promotions, store profile	Yes, per the beta guide
Docs MCP Server	Live	Searches BigCommerce documentation only	No

The practical point: if you searched for a "BigCommerce MCP server" hoping to ask Claude about last month's orders or bulk-edit prices, the official product for that is the closed beta. The live products serve shoppers and developers.

What does BigCommerce Storefront MCP do?

BigCommerce Storefront MCP lets an AI agent act like a shopper on your storefront. According to the official docs, agents can search products, manage carts and hand off to checkout using natural language. BigCommerce's blog adds that agents can retrieve product details and variants and generate a checkout URL.

Key details from the documentation:

• Each storefront has its own unique MCP endpoint URL, shown in the control panel.
• The B2C server supports guest shopping. No customer login is required. An agent can search the catalog, build a cart and produce a checkout link.
• The search_products tool requires a search term of at least 3 characters and supports cursor pagination (paging through results in chunks).
• An optional context field must not contain personal data (PII).
• Documented client setups are Claude Desktop, Claude Code, Cursor and Warp.

Guest carts on a BigCommerce storefront are capped at 50 items, according to the platform limits page. That matters if you test with an agent that builds large carts.

How do I turn on Storefront MCP?

1. Sign in to the BigCommerce control panel.
2. Open the MCP page and switch on the Storefront MCP toggle, then accept the terms of service.
3. Copy the endpoint URL for the storefront you want to expose.
4. Add that URL as a remote server in a supported client (Claude Desktop, Claude Code, Cursor or Warp), following the client steps in the docs.
5. Test with a plain request such as "find products matching a term and add one to a cart," then check the checkout link manually.

BigCommerce's blog describes enabling it under Settings, then Early Access, while the docs describe an MCP page. The blog title says "Now Live," but one third-party review calls it open beta and dates the launch to May 11, 2026, and a financial news summary of the April 2026 Commerce Live event says it was rolled out to all stores. These sources do not fully agree on the date or on whether it is general availability or beta. Check the label in your own control panel.

What can Storefront MCP not do?

• It is not an admin tool. It does not edit products, change prices or read orders.
• It does not currently cover logged-in shopping such as personalized pricing or order history. A secondary source says authenticated shopping is on the roadmap, but I could not confirm that in BigCommerce documentation.
• It does not complete payment on its own. It hands the shopper a checkout link.
• I did not find a documented B2B equivalent of the B2C page. Check the docs navigation if you sell B2B.

What is the BigCommerce Store Management MCP?

This is the server most merchants actually want. Per BigCommerce's beta guide, it lets an AI agent (Claude, Cursor, ChatGPT or any MCP client) work with catalog, pricing, inventory, orders, promotions and store profile. The docs describe tasks such as managing catalog and promotions, looking up and refunding orders, and reading store details.

The page is labeled "Closed beta developer guide," so it is not generally available. I could not retrieve the access process, tool list or authentication model. If you want it, ask your BigCommerce account contact or read the beta page directly, and do not assume it will be available on your plan.

Note that refunds are listed among its capabilities. Any server that can refund orders is a write-access tool and should be treated with care. See MCP Permissions: Read-Only vs Read-Write for Ads and Stores.

What is the BigCommerce Docs MCP Server for?

It is for people building on BigCommerce, such as developers and agencies using a coding assistant. It searches BigCommerce documentation so the assistant gives more accurate answers about APIs. BigCommerce states plainly that it cannot read or change your store, orders, catalog or account settings.

The server address is https://docs.bigcommerce.com/_mcp/server. For Claude Code, the documented command is:

claude mcp add --transport http docs-bigcommerce-com https://docs.bigcommerce.com/_mcp/server

Codex and Cursor setups are also documented. The docs additionally publish /llms.txt indexes and Markdown versions of pages, which AI tools can read without MCP.

What can I use instead for BigCommerce admin tasks?

If you cannot access the closed beta, you have these options. None is endorsed by BigCommerce unless stated.

Need	Option	Main caveat
Shopper-facing agent	Official Storefront MCP	Guest shopping only, per current docs
Admin work, official	Store Management MCP	Closed beta
Admin work, direct	BigCommerce REST and GraphQL APIs	Needs a developer, no chat interface by itself
Admin work, managed	Managed connectors such as StackOne, Pipedream or CData Connect AI	Third party holds access to your store, check pricing and data handling
Admin work, self-hosted	Community servers or a custom gateway	You own the security review
Coding help	Official Docs MCP, or the community @erplinker/bigcommerce-mcp server	Documentation only

Community servers

Several open-source servers exist, and they vary a lot:

• @erplinker/bigcommerce-mcp is a local server with 12 read-only tools that covers documentation and API reference, not store access.
• isaacgounton/bigcommerce-api-mcp wraps the REST API using a store hash and API token. It has read tools for products, customers, orders, categories and brands, and can make changes when you enable that. It is a small project, so review the code before trusting it.
• CData offers a local read-only server over JDBC (a database-style connector). Full create, update and delete work is pointed to its managed CData Connect AI platform.
• A "writable" community server appears in directories, but I only saw a listing, so treat it as unverified.

Managed connectors

StackOne and Pipedream both list BigCommerce MCP connectors. These run on the vendor's infrastructure, which is convenient but means a third party stores credentials for your store. Verify tool lists, permissions and pricing in each vendor's own documentation.

A custom gateway

Some agencies build a small MCP server that signs in with a store-level API account and exposes only a few named tools. The sensible approach is to start with read-only tools, limit scopes on the API account, and add rate limiting. This is a design opinion from a vendor blog, not a BigCommerce standard, but it matches general MCP security practice. The MCP Security: Is MCP Safe for Your Store? A Merchant Checklist covers the questions to ask.

Is MCP how I sell through ChatGPT, Gemini or Copilot?

No. This is a common mix-up. MCP connects an assistant to tools and data. Selling products inside AI shopping surfaces uses separate protocols and feeds.

BigCommerce's parent company, now branded Commerce, announced at Commerce Live in April 2026 distribution across AI discovery surfaces including ChatGPT, Gemini, Microsoft Copilot, Perplexity, PayPal and Stripe, plus agent-enabled checkout and an AI merchant assistant called BigCommerce Companion. The excerpts I reviewed do not make clear which of these are generally available and which are roadmap.

Related items:

• Stripe's Agentic Commerce Suite integration was announced in December 2025 in forward-looking wording. Check its current availability.
• Google agentic checkout through UCP (Universal Commerce Protocol) is described as early access through the Commerce AI Labs program, with timing of early Q4 2026 subject to Google approval.
• Feedonomics and Channel Manager are the product-feed side of this work.

For how these protocols differ, see Agentic Commerce Protocol Compared: ACP, UCP, AP2 and MCP.

What can go wrong with BigCommerce MCP?

• Wrong server for the job. Merchants connect the docs server and expect store data. It has none.
• Over-permissioned API tokens. A token with full write scope behind a community server lets an assistant edit prices or refund orders. Create a token with read scopes only first.
• Unreviewed community code. Small open-source projects have few users and no support commitments. Read the code or run it in a test store.
• Status drift. Beta labels, enablement paths and availability are changing. Do not publish plans based on a blog headline alone.
• Personal data in prompts. The Storefront context field must not contain PII, so do not paste customer details into agent requests.

A safe order of work: test on a sandbox or staging store, begin read-only, check results against the control panel, and only then consider write tools.