Security & Permissions

Authentication

The connector uses OAuth 2.1 with PKCE. You log in with your own Paxper account — there’s no separate API key or service account to manage. The connection is bound to a single organization, chosen when you approve access at the consent screen.

Permissions

The assistant acts as you, with the same role-based permissions you have in the Paxper dashboard:

A Viewer connected via MCP gets read-only access — read tools work, write tools are rejected.
Other roles get the same read/write access through MCP that they have in the UI.
If your organization’s subscription is expired or past due, the connection is read-only regardless of role, the same as the dashboard.

Removed from the organization? Your MCP connection loses access immediately, the same as your dashboard session.

Tokens

Access tokens last 1 hour.
Refresh tokens last 30 days and rotate on every use. If a refresh token is ever reused after rotation (a sign it may have been copied or intercepted), Paxper revokes the connection.

Secrets

Secret values (passwords, keys, tokens, and similar fields) are redacted in every tool’s output, including get_resource. The assistant can tell a field is set without ever seeing its value.

IP allowlisting

If your organization restricts access by IP allowlist, that allowlist applies to MCP requests too:

claude.ai connects from Anthropic’s IP ranges, not your own — if your organization uses an IP allowlist, you’ll need to allow Anthropic’s ranges for the connector to work from claude.ai.
Local clients (Claude Desktop, Claude Code, and similar tools running on your machine) connect from your own IP, so they work with an IP allowlist the same way your browser does.

Disconnecting

You can revoke your own connection at any time from Settings → Connected Apps. Organization admins can revoke any member’s connection from the same page.