13.3What an agent can do

The tools

The server offers 95 tools, grouped by area below. Tools that only read are marked read-only to the client; clients may run those without asking you first. Which tools an agent can actually use depends on its permissions and on whether it has read-only or read-and-write access.

Users

Groups

Computers

Accounts

Organizational units

Any directory object

Password policies

Security and delegation

Service accounts

Domain and forest information

Saved searches and operations

The exact list and wording an agent sees is on the MCP Tools screen. Some capabilities are deliberately not offered to agents at all, for example the REST API's multi-object batch requests, raw access control entry editing and approving operations.

When a change waits for approval

When an agent's change has to wait for a person, the tool does not perform it. Instead it answers with a pending result such as:

{
  "status": "pendingApproval",
  "code": "Operation.ApprovalRequired",
  "operationId": "8d1f6c2e-4b5a-4e8f-9a37-0c2d5e6f7a81",
  "displayName": "Delete User",
  "expiresAt": "2026-10-04T09:30:00Z",
  "canApprove": false,
  "reason": "Level",
  "reasonDetail": "This agent asks before destructive and security changes.",
  "message": "… Nothing has been changed yet. Do not call this tool again for the same change: call GetAdOperationStatus with this operationId …"
}

What the fields mean:

• operationId — the handle for the queued operation.
• reason and reasonDetail — why it is waiting: a protected target (ProtectedTarget), an operation pinned to always ask agents (Pinned), the agent's supervision level (Level), or an unusual number of recent destructive changes by the same credential (Volume). They call for different answers — wait, change the request, or ask an administrator about the agent's level.
• canApprove — whether the person the agent acts for may approve it. If they may, and may also open the operations log, an approvalUrl links to the operation in the console, where they sign in and decide. The agent itself can never approve.

There is deliberately no success field: the change neither failed nor succeeded yet, and retrying would put a second request in front of the approver.

The agent follows up with GetAdOperationStatus and the operationId. Its status is WaitingToApprove until somebody decides, then Success, Rejected, Cancelled (the requester withdrew it) or Failed (which includes a request that expired undecided). Once the change has run, a result block names the object it affected; for a create, result.entityId is the new object's GUID. An agent can only read the status of operations its own person requested. The tool needs operations:waiting:list.

Approvers see the queued operation, the agent's name and the reason in the operations log and in their notifications (see Chapter 9, Operations log, approvals and waiting operations).

Your connected agents

Every person can see and disconnect the agents they connected.

1. 1Open My profile from the menu under your name (the page is at /dashboard/profile).
2. 2Go to the Connected agents section.

Each agent shows its access (Read only or Read and write), the role and directory you chose, when it was connected, when it was last used and when it expires. Select Disconnect to stop an agent; it stops on its next request. You can connect it again at any time from the agent.