AI agent account

What an agent account is in SAQ, how an Owner or Admin creates one, its scopes and keys, how its actions are attributed, and what it can never do.

Written for
admin, integration-developer
Roles
agent, owner, admin
Requires
Owner or Admin with a recent full login and two-factor authentication, to create or rotate
Feature
tokens

An agent account is a workspace user that exists only for an AI assistant or an integration. It has no password and no login; it acts through one API key. Where a personal token acts on behalf of a person, an agent's actions are attributed to the agent itself, which keeps a person's name off changes they did not make.

What an agent is

  • A token-only workspace user with the Member role. The role cannot be changed ("Agent accounts stay workspace members.").
  • Addressed as handle@slug, for example claude@acme. The address is not a mailbox; nothing is ever sent to it.
  • A member of every project, including projects created later, so its key sees what a Member on every project sees: all three comment audiences, estimates, notes, and non-billable time, like any workspace user.
  • Free of charge: agent accounts do not count against the plan's seats.
  • Holder of exactly one live key at a time.

Create an agent

You need the Owner or Admin role, a recent full login, and two-factor authentication.

  1. Open Settings → Members and go to Agent accounts.
  2. Enter a Name (how the agent appears in timelines) and a Handle ("The agent's address becomes handle@{slug}."). Handles are lowercase letters, digits, dots, hyphens, and underscores.
  3. Choose the Scopes and, optionally, allowed IP addresses.
  4. Select Create agent. The key is shown once: "Agent created. Copy the key now."

Put the key into the assistant's configuration; Settings → MCP setup has ready-made snippets for Claude Code and Cursor. See Connect an AI agent.

Scopes

Scope What it allows
Read Always included: tickets, boards, My work, projects, time, documents the agent may see
Write Create and update tickets, comment, link, log time, move tickets
Triage (shared Inbox) See the Inbox and route or merge messages

An agent key is never restricted to a project list; project scoping is a feature of personal tokens.

Rotate and revoke

On the Agent accounts table each row shows the Address and the key Prefix.

  • Rotate revokes the current key and issues a new one with the same name, scopes, allowed addresses, and expiry: "The current key stops working and the new one is shown once."
  • Revoke stops the key at once: "The agent loses access immediately. A new key can be created later." A later Rotate restores the previous key's settings; an agent that never had a key must be issued one with explicit scopes.

Revoked keys stay in the token registry under Settings → API tokens for 30 days. See API tokens and agents.

Attribution

  • In ticket timelines and comments the agent appears under its name, marked (agent) in the member list.
  • Tickets and comments created through the API carry the source api; through the MCP server, mcp.
  • The audit log records the agent as the actor. When a person's own token is used instead, the log records the person, with a note that a token acted on their behalf.

What an agent cannot do

  • Log in to the app, use a magic link, or sign in through single sign-on ("Agent accounts cannot log in.").
  • Create, list, rotate, or revoke tokens, including its own.
  • Close a period or create or void a billing run; these refuse every token.
  • Manage clients, workspace settings, workflow, mailbox, or authentication; these are browser-only.
  • Archive or delete tickets, add or remove viewers, or read account pages; these are not on the token route list.
  • Receive email; the agent's notification setting is fixed to none.

For the exact tool list and request shapes see MCP server.