MCP & Approval Mode

Settings → MCP controls the default IDE approval mode for the tenant — how much human-in-the-loop confirmation VibeReview requires before applying guardrails in the AI IDE.

Default IDE approval mode

This setting is the tenant-wide default for every project. Individual projects can override it in their own Settings tab; individual or single-project tenants use this as the only knob.

Mode	Behavior
Auto	Apply guardrails silently (default).
Critical only	Ask before authentication / authorization / crypto / secrets changes.
Always	Ask before every guardrail change.

Click Save default to apply.

Critical categories

When the mode is critical or always, guardrails in these categories are tagged requires_approval=true, so the AI assistant pauses for human confirmation before applying them:

• authentication
• authorization
• data_exposure
• data_integrity

How it maps to the CLI

The tenant/project approval mode corresponds to the CLI's --approval-mode flag on init, whose values are auto, critical, and always. Setting it here establishes the default; the CLI flag can set or override it locally for a given workspace.

Choosing a mode

You want…	Use
Frictionless secure-by-default coding	Auto
A checkpoint on the highest-risk changes only	Critical only
Maximum oversight / regulated environments	Always

Start at Critical

Critical only is a good balance for most teams — guardrails apply automatically for routine work, while sensitive changes to auth, authorization, secrets, and data exposure still get a human in the loop.