Tokens

Personal access tokens authenticate the MCP server and the API. Every call made with your token is attributed to you — and, for an agent's token, to the agent acting through you.

Creating one

Give it a label, optionally an expiry, and say:

• Used by — me, or a script that isn't an AI agent, or an AI agent. An agent's token can propose but never approve: it can draft responses and broadcasts and edit its own drafts, but it can't send or approve a response, send a broadcast, approve a library item, or touch users, entities or anyone's credentials — regardless of the access you grant it. Its drafts always wait for a person's review.
• Access by area — for each area (cases, correspondence, time, library, broadcasts, messages, reporting, administration) choose none, read only or full. A script that only reads the dashboard gets read on reporting and nothing else. Messages (the log of every email and text, and re-routing one) is only for an OPN admin's token.

The secret is shown once, at creation. Copy it into your .mcp.json then; it can't be shown again.

What a token can do

Its reach is the grants × the agent cap × your current role, worked out on every call. Levels your role can't hold are greyed out in the form, and the server refuses them too. If your role changes, the token changes with it — there's nothing to reissue. A junior operator's token can't send responses, because a junior can't; an agent's token can't, whoever owns it.

Revoking

Revoke stops a token immediately. Admins see every user's tokens (labels, use and access, never secrets) under User tokens, and can revoke them there.