What Is an MCP Server for Ecommerce? A Plain Guide

An MCP server for ecommerce is a service that lets an AI app, such as Claude, ChatGPT or Cursor, use your store's real actions: list orders, create a product, make a discount code, build a checkout link. Pocketsflow runs one at https://www.pocketsflow.com/mcp, with 303 tools in 48 groups that cover a whole digital product store.

What MCP is, in plain words

MCP stands for Model Context Protocol. The official site, modelcontextprotocol.io, calls it an open standard for connecting AI applications to outside systems. Three parts talk to each other: the host is the AI app you chat in, the client is the connector inside that app, and the server is the service on the other side, such as Pocketsflow. They exchange JSON-RPC 2.0 messages.

The MCP specification (version 2026-07-28, read November 1, 2026) lets a server offer three kinds of things: resources (data to read), prompts (message templates) and tools (functions the AI can run). Pocketsflow's server offers tools only. Its resource and prompt lists answer empty, because every store action is a tool.

What a tool is on an ecommerce MCP server

A tool is one named action with a description and a list of inputs. When you ask "make a 20% code for my course", the AI app reads Pocketsflow's tool list, picks create_discount, fills in the code, the value and the product, and calls it. Pocketsflow runs the same public API a developer would call, then returns the result for the AI to read back to you.

The MCP spec says the AI app must get your consent before it runs a tool, so you stay in the loop. Pocketsflow's setup prompt also asks the AI to check with you before anything goes live.

How OAuth sign-in works on the Pocketsflow MCP server

Pocketsflow uses OAuth 2.1 with PKCE, the sign-in standard the MCP spec names, so you never paste a password or an API key into the AI. The setup is the same in every app that supports it:

1. Add https://www.pocketsflow.com/mcp to your AI app as a custom connector or remote MCP server. Leave any token or header fields empty.
2. Your browser opens the Pocketsflow consent page. Sign in, and check which app is asking and which account it will use.
3. Click Allow access. Pocketsflow creates a key named "MCP: <app>" in the mode your dashboard is in, live or test.
4. Ask: "Show me what you can do in my Pocketsflow store."

The AI app only gets short tokens: access lasts one hour, and the refresh token rotates on every use. To disconnect, delete the "MCP: <app>" key under Developers, API keys. For a server or CI job with no browser, Pocketsflow also accepts an API key through the @pocketsflow/mcp package.

What Pocketsflow's ecommerce MCP server exposes

Pocketsflow's MCP server mirrors the store, group by group: products, checkout, discounts, orders, subscriptions, memberships, newsletters, analytics and more. The chart shows the eight largest groups. Each bar pair splits a group into look-up tools (names that start with get_, list_ or search_) and tools that change something.

Memberships and communities is the largest group, with 17 tools. Across the whole server, 158 tools look things up and 145 change something. Look-up tools are the safe place to start: "how did sales go this week?" only reads.

Which Pocketsflow MCP actions are read only

Pocketsflow keeps money movement out of the AI. The payouts group is read only from end to end: 4 tools (get_payout_status, get_payout_balance, list_payouts, get_payout) answer "how much can I withdraw?" and "did my payout land?", and nothing more.

The refund tools are easy to misread. create_refund, list_refunds and get_refund manage your refund policies, the rules shown to buyers. They never refund a payment.

What stays in the Pocketsflow dashboard

• Withdrawing money, and adding or changing a bank account.
• Identity checks and information requests for payouts.
• Refunding a payment to a buyer.
• Uploading images: file uploads are not part of the tool list.
• Upgrading your plan.

Paid features stay behind the same plan gates as the dashboard. Upsells, abandoned checkout emails and affiliates need Growth; memberships, the customer CRM and A/B tests need Pro. The server tells the AI to call get_entitlements first, and a FEATURE_LOCKED error means the AI should send you to billing, not retry.

Prompts to try on the Pocketsflow MCP server

• "How did sales go this week, and who are my top customers?" calls list_orders and list_customers.
• "Create a $29 presets pack and give me the checkout link." calls create_product and create_checkout_session.
• "Make SPRING20: 20% off everything until Sunday." calls create_discount.
• "Write a newsletter about my spring sale and send it." calls create_post and send_post.
• "Are my payouts active, and how much can I withdraw?" calls get_payout_status and get_payout_balance.

Switch the dashboard to test mode before you connect to try these with no real money.

When an MCP server is the wrong fit

An MCP server is for a person talking to an AI. If your app must create a checkout for every signup with no one in the loop, call the Pocketsflow API or the pocketsflow SDK from your code instead, and listen to webhooks. The trade-offs are in MCP vs API for selling online.

Start with the Pocketsflow MCP server

Connecting costs nothing extra: the MCP server is part of every Pocketsflow plan, including the free one. The Pocketsflow MCP page has the setup for each AI app. For the full tool tour, read what Pocketsflow MCP can do, and for the bigger picture, read what agentic commerce is.