Checklist — Platform MCP (Shopify first)

The Score rates this checkpoint from two questions: whether an AI assistant can read and update your store directly, and how complete your catalog data is. This is what each rating looks like in practice.

0 or 1: not started, or ad hoc

• Everything goes through the admin or a developer. At most, CSVs get exported and pasted into a chat now and then.
• Catalog data is patchy: lots of gaps and duplicates, or fine on the best sellers and thin everywhere else.

A 2: in place, not connected to AI

• Your store data is clean and structured, but no AI is connected to it.
• The catalog is complete and consistent: titles, descriptions, attributes, images and custom fields.
• It stays that way because someone maintains it by hand.
• Every custom field has a known purpose and a known owner.

A 3: connected, used sometimes

Everything in a 2, plus:

• Claude can read and update your store through an MCP server or API connection.
• You use it for some jobs, like a catalog clean-up or a bulk edit.
• AI fills or fixes gaps in the catalog when you ask it to.
• You’ve decided what the connection can change, and it can’t change anything else.

A 4: connected, and part of how work gets done

Everything in a 3, plus:

• Routine catalog and merchandising work happens through Claude, not the admin.
• AI keeps the catalog complete as products are added, so gaps don’t build up.
• Changes that need a human sign-off get one before they go live.
• You can see what was changed, and when.

You’ve earned a 4 when…

• Adding a product, fixing a range or building a collection starts as a conversation.
• Nobody exports a CSV to do a job the connection can do.
• New products arrive complete, not “we’ll fill it in later”.
• You trust the connection with your live store, because you decided what it can touch and you can see what it did.

Next track: Marketing MCP (Klaviyo).