Skip to main content
Retailgrid exposes an MCP endpoint, so an AI assistant can read your catalog and your grids and answer questions about them in your own words — no exporting. This page walks through connecting a client. For what the connection can and can’t do, and how to manage or disconnect clients, see AI connections.
AI connections is rolled out per organization. If it isn’t available to you yet, ask us to switch it on and we’ll send you the endpoint to use.

​
Connect a client

The connection starts in your AI client, not in Retailgrid. In broad terms:
  1. Add Retailgrid as an MCP server in your client, using the endpoint we provide for your organization.
  2. Sign in. The client sends you to Retailgrid to sign in if you aren’t already.
  3. Approve the consent screen. Retailgrid shows you which client is asking for access. Read it, then approve or decline. Because consent is tied to your own sign-in, the client only ever gets the access you already have — it can’t see your teammates’ grids.
  4. The client connects and appears in the list on the AI connections page.
The exact steps for adding an MCP server differ by client — follow your client’s instructions for Claude, ChatGPT, Cursor, or another MCP-capable tool, using the Retailgrid endpoint at the server-URL step.

​
Try three questions first

Once connected, confirm it works with a few plain-language questions about your own data:
  • “Show me all overpriced SKUs.”
  • “What is the CPI min price for Category X?”
  • “List my KVI products.”
The assistant answers from the same data and permissions you’d see in the grid.

​
Good to know

  • The connection is read-only — an assistant can look at your data and reason about it, but it can’t change a price, run an agent, or write anything back.
  • A client sees the grids you own, not your teammates’.
  • For scheduled syncs into an ERP or BI tool, use the public API instead — it’s built for extraction, paginated, and scoped to an organization key rather than one person.