Connector access and security checks

You can now choose which apps each connected assistant can work on. New connections start at All apps, including shared apps and apps you add later. Your existing permissions still apply. Choose Only selected apps to limit access; that connection can update those apps but cannot create new ones.

Database tools start at None. For apps you own, you can choose Read or Read and write, then choose whether the assistant acts as a visitor, an app user or with full access. The app must also be included in the connection's app access. Turning these tools off does not stop the app itself from using its database.

Change or revoke access under Account → Connections. Existing connections keep their settings. Changes apply to new requests; work already started may finish.

Before publishing, Percher now blocks critical vulnerabilities and those listed as being exploited in attacks. It also stops if the security check cannot finish. The message explains what needs to happen: update something in your app, wait for a Percher fix, or retry when the check is available. Passing a scan does not mean every vulnerability has been found. See security scanning.

Update to CLI 1.4.0 or local MCP 1.3.0 for these explanations. Hosted connectors receive them without a local update.

Also changed:

• Emails use the same Percher layout. You can turn optional emails off in notification settings. Required account and security notices cannot be turned off there.
• The status page refreshes automatically and shows Stockholm time with its UTC offset. Old readings no longer show as operational.
• The CLI explains why you cannot publish anonymously while signed in.