Some MCP Hub actions wait for a person before the assistant can carry them out. Reviewing the pending request in the module is a separate step from approving a tool call in the assistant’s own interface.

Find the request

Open PD Modules → MCP Hub → Approvals. Read What it wants to do, What it had just read and Expires together. Match the intended operation and target to your actual request, rather than relying on the assistant’s conversational summary.

The hub builds the approval description from the submitted operation. The recent-read context can help you notice a request that followed unexpected material, but it is not a complete conversation transcript or proof that the proposal is appropriate.

Decide on this operation

1. Check the exact target, requested change and any consequence shown in the summary.
2. Review the recent-read context and the request’s remaining lifetime.
3. Choose Reject if the proposed operation is not the one you intended.
4. Choose Approve only for the operation you have reviewed, then read and confirm the approval prompt.
5. Let the assistant resume that approved request and check the actual result.

Approval does not itself execute the tool. The assistant must return to carry out the approved operation; it can still fail another access or execution check. Changing the conversation or submitting different arguments is not approval of a different operation.

Verify before any retry

Inspect Audit trail and the affected item in your normal admin screen. A message saying Approved does not mean the change has already happened. Do not submit the operation repeatedly because an assistant has not yet shown a result.

Pending approvals normally expire after one hour. If a request has expired or been rejected, stop that attempt. For an operation you still need, first reconcile any earlier outcome, then make a fresh explicit request and review its new approval.

To require a human decision for ordinary writes too, edit the intended connection and enable Ask a person before anything that changes data, not just the risky ones. This does not raise its ceiling or staff permissions, and reads are not queued for approval.

See the PD MCP Hub Blesta guide for related tasks, or explore PD MCP Hub.