You add an allow rule for a particular MCP tool and switch to a permissive mode so the session runs unattended. That one tool still stops and asks for approval every time it is called. Why?

The principle — Some approvals are the product rather than an obstacle to it. Where a tool exists to obtain a person's agreement, automating the agreement would destroy the thing it is for.

Why the key is correct — When a server marks a tool as requiring user interaction, Claude Code shows that tool's permission prompt on every call, even in the permissive modes, and does not offer an option to stop asking. Allow rules that match the tool do not skip the prompt either.

Why the others are wrong — The rule is not written wrongly; it is simply outranked. Permissive modes do have this exception. And the prompt is not repeating because a choice was lost, since no option to save one is offered.

Remember this — A tool whose prompt is the point cannot be pre-approved, and the absence of a remember-my-choice option is how you recognise one.

Sources — Anthropic's Claude Code MCP reference.