Skip to main content
Any client that can send an Authorization header can use the connector. These skip the browser step entirely — the token travels in the header.
Copy these from the app, not from here. In the Access tokens table, Copy config copies a ready config for that token, with a <YOUR_TOKEN> placeholder where your token goes. The shapes below are for reference.
For clients that read an mcp.json-style server list:

​
No connector dialog in your app?

The AI connectors (MCP) tab has a card for exactly this: No connector dialog in your app? Press Show config to see the same three shapes — Claude / Cursor / VS Code, Claude Desktop (older) and Terminal — and Hide config to close it.

​
Two things to get right

The store comes from the token. The URL the app prints may include ?shop=; it only tells two stores’ connectors apart, so the config works with or without it. Name the server entry per token. A config file is keyed by that name: two entries both called notifylyst means the second silently replaces the first. The app names each one notifylyst- plus the token’s name (notifylyst-claude-desktop) for exactly this reason.

​
Checking a token without starting a session

The connector has a health check that reports what a token allows — useful when a client is being unhelpful about why it failed:
A good token returns status ready with the store, the allowed tools and the expiry. Any refusal returns 401 (or 429 when you are over the rate limit). Each check uses one request from your rate limit. Without a token it is a plain liveness check.
If you monitor uptime, point the monitor at /healthz — never at /mcp, which answers 405 to a GET by design and will read as unhealthy. See Error codes.

​
Next

Try an example — three things worth asking on day one.