Connecting an AI assistant
How to let an AI assistant help manage your server through a scoped API token, with confirmation for sensitive actions.
You can connect an AI assistant to your JSOPanel server using the same kind of API token used for billing integrations, letting you or your team ask it to carry out routine tasks, checking an account's status, looking up information, or making a straightforward change, through natural conversation instead of clicking through the panel yourself.
Setting up the connection
- Create a new API token specifically for this purpose, and tick only the permissions you want the assistant to have. See the article on creating API tokens for how.
- Give your assistant your server's address and this token, following that assistant's own instructions for connecting to an external service. It authenticates the same way any API token does, as a bearer token on every request.
Connecting Claude specifically
JSOPanel's assistant connection speaks MCP (Model Context Protocol), which Claude supports as a custom connector, in both Claude.ai in a browser and the Claude desktop app. The exact wording of Claude's own menus can change over time, but as of this writing:
- In Claude, go to Customize → Connectors (on a Team or Enterprise plan, an Owner adds it first from Organization settings → Connectors, and everyone else then connects to it from Customize → Connectors).
- Click Add custom connector.
- Give it a name, and for the server URL enter your server's address followed by
/mcp, for examplehttps://yourpanel.example.com/mcp. - For Authentication, choose No sign-in, JSOPanel doesn't use Claude's own OAuth sign-in flow for this.
- Open Request headers. Add one: pick authorization from the list (or choose Custom header and type it), and for the value enter
Bearerfollowed immediately by the token you copied when creating it, for exampleBearer jsop_abc123.... Mark it Required. - Click Add to finish, then turn the connector on for a conversation from the + button in the chat box, under Connectors.
How permissions work here
Exactly as with any other API token, the assistant can only do what the token allows. If you only grant read access, it can answer questions about your accounts but can't change anything. If you grant it broader permissions, it can take action within those limits.
Confirmation for the most sensitive actions
Three actions get an extra safeguard: deleting an account, deleting a domain, and reading a mailbox's messages. On the MCP & API tokens page, under AI assistant permissions, choose whether you want to be asked to confirm one of these before it happens, or simply told after it's already been done. Either way, you're always told.
With "Ask me to confirm first" selected, the assistant's request is held and you're sent a message with the details, and the action only goes ahead once you approve it. This lets you give an assistant broader permissions for convenience, while keeping a manual check in place for the actions that matter most. This setting applies to your own account; each account that connects an assistant chooses it for themselves.
Reviewing what the assistant has done
Actions taken through the token are visible the same way any other API activity is, from your MCP & API tokens page and your server's audit log, so you always have a clear record of what was done and when.