Security and scopes
What an agent connected over MCP can see, what it must be granted before it spends or sends, and how to take that grant away.
Everything here applies to the connector at https://mcp.argorant.com/mcp.
Boundaries
- A connector is bound to one signed-in Argorant account. Nothing is anonymous; every call is attributable.
- Counts, coverage and masked previews are free. Preview rows never carry emails, phones or profile URLs.
- Reveals, enrichment and exports run against the connected workspace's plan and credit balance.
- Campaign tools act on the connected workspace only. Launching and replying send real email; those tools are annotated as destructive and open-world so clients ask for confirmation.
- Workspaces are isolated: an agent cannot list or touch another workspace's campaigns, leads or replies.
Scopes
| Scope | Cost | Unlocks |
|---|---|---|
argorant:read_counts | Free | Counts, account state, campaign and inbox reads |
argorant:search_segments | Free | Coverage search and masked previews |
argorant:manage_lists | Free | Saved lists; campaign writes, launch, replies, reply statuses |
argorant:create_exports | Credits | Export jobs and list exports |
argorant:unlock_contacts | Credits | Reveal and person enrichment |
Tool annotations
Every tool declares readOnlyHint, openWorldHint and destructiveHint. Read tools update private usage records and are therefore not marked read-only; argorant_launch_campaign, argorant_inbox_reply, argorant_delete_campaign, argorant_set_campaign_emails and argorant_update_campaign are destructive; anything that probes mail servers or sends email is open-world.
Revoking access
Disconnect the connector in the client, or revoke the grant at app.argorant.com/profile → Connected apps. Rotating or deleting an API key does not affect MCP grants, and vice versa.