Skip to main content
The short path. Every option is explained in Setup if you want the detail — this page is the fastest route to a working connection.

Before you start

  • Notifylyst installed on your Shopify store.
  • A plan that includes AI MCP Integration. Without it you can still authorize, but every tool call is refused with MCP_PLAN_DISABLED.
  • Claude (web, desktop or mobile). Other clients work too — see Other AI clients.
1

Create a read-only token

In Notifylyst, open Settings → Developers → AI connectors (MCP). Check the Connector card shows On, then press Create token:
  1. Give it a Token name after the client — “Claude Desktop”. You will thank yourself in six months when deciding what to revoke.
  2. Under What it can do, choose Read only. That grants the nine read tools and none of the seven that change anything — the right way to start.
  3. Leave Expires on Never expires unless the access is temporary.
  4. Copy the token now. It starts with nl_mcp_ and is shown once, in the banner right after creation.
Lost it before copying? You don’t need to start over — find the token in Access tokens and press Regenerate. That issues a new secret for the same token, keeping its name, tools and expiry, and the old one stops working.
2

Add the connector in Claude

Settings → Connectors → Add custom connector, and paste your connector URL:
Copy the real URL from the Connector URL field rather than typing it. Make sure it ends in /mcp (plus the ?shop= part the app adds).
Press Connect. A browser page opens:
  1. Connect to Notifylyst — paste the nl_mcp_ token, press Continue.
  2. Authorize access — check the store and the tools are right, press Authorize.
You go straight back to Claude, which now shows the connector as connected.
3

Ask your first question

Start a new chat:
You should get your store details, plan, active channels and wallet balance back.Then try the one that usually pays for the setup time:
That chains your waiting-list demand to a live Shopify inventory lookup — see Find restock opportunities for what a good answer looks like.

Something went wrong?

Troubleshooting is a symptom → cause → fix table, and it covers the errors that look alarming but are working as designed.