---
title: "What Is an MCP Server? Model Context Protocol | SKU.io"
url: "https://www.sku.io/glossary/mcp-server/"
description: "An MCP server lets an AI assistant like Claude or ChatGPT use a business system's data and actions. How it works, a worked example, and what to check first."
published: 2026-11-21
---

Ecommerce operations glossary

# What is an MCP server?

Also called Model Context Protocol server, MCP connector

An MCP server is a service that lets an AI assistant use another system's data and actions through the Model Context Protocol, an open standard. The server publishes a set of tools, such as look up stock or draft a purchase order, and the assistant calls them during a conversation, within the permissions the user granted when connecting.

## Worked example

A buyer asks an assistant connected to the inventory system’s MCP server: “Do we have enough of the 750 ml bottle to cover the next two weeks?”

The assistant picks two tools from the server and calls them:

1.  **Stock by warehouse:** 120 units in the main warehouse, 45 at the 3PL, so 120 + 45 = 165 on hand.
2.  **Sales and inbound:** 15 units a day over the last 30 days, and 300 units on a purchase order due in 9 days.

Then it does the arithmetic:

-   165 ÷ 15 = **11 days** of cover, so the stock lasts past the 9 days to the delivery
-   Stock when the order lands: 165 − (9 × 15) = 30 units, then 330 after receiving

Its answer: yes, with 30 units to spare on the day the purchase order arrives. Every figure came from the system, not from the assistant’s memory, and the buyer can open each record it cites.

## MCP vs an API

An API is built for code: a developer reads the reference, writes the calls and handles the errors. An MCP server is built for an assistant: each tool comes with a description the model reads, so a person can ask in plain words and the assistant chooses the calls. Many MCP servers sit on top of the same API, with the same permissions underneath.

The trade-off is cost per conversation. The assistant loads the tool descriptions before it starts, so for a job that runs every night, a script against the API is cheaper and more predictable.

## What to check before you connect one

-   Does it start read only, with changes granted area by area?
-   Does it create drafts for anything that commits money or stock?
-   Can it delete records? It should not need to.
-   Is every call logged, with who made it and what changed?
-   Can an admin switch it off for the whole account at once?

## In SKU.io

SKU.io runs an MCP server with one connector URL for Claude, ChatGPT or Cursor. You choose which areas it may change when you connect, the sales and purchase orders it creates are drafts, and every call is listed on an Activity tab for 180 days.

[How it works, in the SKU.io docs](https://www.sku.io/docs/guides/settings/developer/apps-and-webhooks/connect-an-ai-assistant/)

## Read next

-   [Running inventory with AI agents](https://www.sku.io/run-inventory-with-ai-agents/)
-   [How to test a vendor's embedded AI assistant](https://www.sku.io/embedded-ai-assistant-vendor-test/)
-   [Agentic commerce](https://www.sku.io/glossary/agentic-commerce/)
-   [AI agents in inventory operations](https://www.sku.io/glossary/ai-agent/)
-   [The SKU.io inventory management API](https://www.sku.io/inventory-management-api/)

## More terms: APIs and AI agents

-   [Agentic commerce](https://www.sku.io/glossary/agentic-commerce/)
-   [AI agent](https://www.sku.io/glossary/ai-agent/)
-   [Inventory management API](https://www.sku.io/glossary/inventory-api/)

Updated November 21, 2026 · [All terms](https://www.sku.io/glossary/)
