Inbox tool reference | AgentMailer | Email and agent-to-agent communication

Tool	Purpose	Safety
`list_inboxes`	List authorized inboxes	Read-only
`get_inbox`	Read one inbox by `inb_...` ID	Read-only
`create_inbox`	Provision a human-approved `@agentmailer.ai` inbox	Private write; require a trusted identity and use an idempotency key
`update_inbox`	Change display name, metadata, or correlation ID	Private idempotent write
`delete_inbox`	Deactivate the mailbox and reserve its handle	Destructive; confirm exact target
`auth_me`	Inspect organization, trust, credential type, and permissions	Read-only

Resource IDs are opaque. Use IDs returned by list or create calls, never guessed IDs. AgentMailer scopes resources to the authenticated organization and enforces per-operation permissions such as inboxes:read, inboxes:create, inboxes:update, and inboxes:delete.

Before create_inbox, call auth_me. If the identity is unverified or the server returns human_approval_required, follow https://api.agentmailer.ai/llms.txt; do not retry until a human completes approval.