Sources checked
What does the tools specification require?
The MCP tools specification says tools are model-controlled: the model may discover and invoke them from the prompt.
It also says that for trust and safety there should always be a human in the loop with the ability to deny tool invocations.
Applications should show which tools are exposed, indicate when a tool is invoked, and present confirmation prompts so a person stays in that loop.
The same page says clients should prompt for confirmation on sensitive operations and show tool inputs before calling the server, to avoid accidental or malicious data exfiltration. The protocol does not design the dialog.
It places the duty on the client that actually performs the call.
Show the exact draft before approval
An email tool is a useful hypothetical. The review should show recipient, subject, body, and attachments, and the approval should bind that draft.
If any of those fields change after the click, the earlier approval does not cover the new action.
Computer-use agents make the same point for a click on a screen, and agent sandboxing is the limit on what code can reach even after a person has said yes to a class of action.
Which calls need the gate?
The spec does not say to confirm every read. It points at sensitive operations and at showing inputs that could send data out.
A practical split is: reads of material the user already opened can stay quiet, while sending, paying, deleting, or changing a record stops for a person.
Enforce that stop in the client that executes the tool, not only as a sentence in the prompt.
Common questions
What is human-in-the-loop AI?
Here it means a person can inspect and deny a model’s tool call before the call runs.
The MCP tools spec uses that phrase as a requirement for tool safety, not as a claim that the model is supervised in general.
Is a system prompt that says “ask first” an approval gate?
No. The gate has to be in the software that performs the action. A prompt can be ignored on the next call.
Sources & further reading
How this story was made
Written by Kristian Kostov with AI assistance and checked against the linked sources. Company performance claims are attributed to the company. Analysis reflects AiLookout’s interpretation; we have not independently tested the products discussed. Cover photography is illustrative and does not depict the specific announcement or product.
Our editorial standards





