Accounts, permissions and revocation

An MCP connection acts through the person who authorizes it. It uses that person's current Insights access and Unify permissions within one connected account. It does not borrow an Eniris administrator's identity.

What you approve

Actual authorization interface with fictitious account data. This demonstration does not create a live connection.

Check three things on this screen:

1. App and callback origin: the assistant you are connecting.
2. Account: the portfolio or organization the connection will use.
3. Editing: whether supported saves should be permitted in addition to reading.

Generating a report or rendering a saved measurement/PV report is classified as a write operation, even when no email is sent. Retrieving an existing PDF is a read operation. Tool reference lists the distinction for each tool.

One connection, one account

The connection is bound to the account displayed during consent. Changing the selected account in a browser tab does not move that connection or give the assistant access to another account. There is no MCP tool to switch accounts.

To work in another account, select it in Unify and complete a new authorization from your AI client. Where the client supports separate connections, use recognizable names for them. Otherwise reconnect the existing entry. Ask portal_guide to confirm the account before starting work.

Two colleagues connecting to the same account can see different sites if their Insights access differs. An organization-wide plugin definition does not grant all colleagues the creator's data permissions: each person must authorize their own access.

Existing permissions remain in effect

The server checks current access for operations and the sites or measurements they reference. Editing permission on one site does not grant editing permission on every site. Account-wide KPI formulas require the account settings permission used by the portal.

If access is removed or an account is disabled, the connector cannot use that old authorization to bypass the change. See Users and access rights for the underlying permission model.

Where data goes

Tool results are sent to the AI service you chose. They can contain site names, measurements, notes, report content and configurations needed for the request. Use your organization's approved AI workspace and its data-handling settings.

Unify does not share your password with the assistant. Credential fields are filtered from returned objects. This does not make site information anonymous: business content that you request still reaches the AI client.

Revoking a connection stops future access through it. It does not erase earlier messages, exported PDFs, screenshots or other data already held by the AI service. Manage those through the service's own retention and deletion controls.

Review and revoke connections

Open MCP in Unify and find My connections. Each entry shows its client, connected account, access mode, callback origin and validity date.

Choose Revoke access for an entry you no longer need. This revokes that authorization; it does not delete dashboards, reports or notes already saved with it. Those changes remain ordinary portal content.

To reduce an existing connection from editing to read only, revoke it and authorize a new read-only connection. Remove unused duplicates so you can recognize the remaining entries.

Session lifetime and reconnecting

The authorization can last at most 30 days and cannot outlive the personal session from which it was approved. Shorter-lived access tokens are refreshed by the client while the grant remains valid. The date shown under My connections is an upper limit, not a guarantee that access survives logout or revocation.

If you sign out, the session expires or the connection is revoked, sign in again and reconnect from the AI client. Do not paste a token or password into the conversation to repair access.

Editing safely

For revision-controlled objects such as dashboards, reports and the KPI formula catalogue, a concurrent edit can cause a conflict. The assistant must read the new version and reconcile your intended change, preserving someone else's work. Repeatedly forcing the old content is not a solution.

Not every editable configuration has the same revision mechanism. Read before writing, limit the change to the requested fields and check the saved result. AI-client approval prompts are an additional control; the server's account and permission checks remain decisive.