Start with the six read-only default tools, add more read tools next, then tools that start scans, and only last tools that change state or touch integrations. Enable one group at a time, watch how assistants use it for a week, and keep a record of what you turned on and why.
Why stage it
Safeguard registers hundreds of MCP tools. Each has its own flag, off by default except for six reads. That design lets you widen access deliberately. Staging turns that into a plan: each step has a reason, an owner and a way back.
Tool gating resolves per tenant, not per user. Everyone in the tenant sees the same set. That makes each decision larger than it looks, so it is worth going slowly.
Stage 0: the default set
The six reads cover products, projects, one project's details, vulnerabilities, one CVE and findings. Nothing in it changes state or touches a credential. Leave it as it is while people learn what assistants can do with it.
Stage 1: more reads
Add tools that read the same kind of data in more depth: SBOMs, components, licenses, compliance reports. Ask for each: could this expose something a viewer should not see in a chat window? Remember that an assistant acts as the signed in user, so Safeguard roles still apply.
Stage 2: tools that start work
Scanning tools start jobs. Turn them on for a pilot group of teams, and tell people. Check that the scan targets are ones you own.
Stage 3: tools that change state
Adding integrations, editing policies, creating tasks, changing members. Enable these only with a clear owner and a clear need. For each, decide who reviews what the assistant did. Prefer a human confirmation step in the client for any call that writes.
Stage 4: billing, account and admin
Keep these off unless you have a specific use. The documentation lists billing, notifications and admin CRUD as opt in for good reason.
A simple decision table
| Question | If yes | If no |
|---|---|---|
| Does the tool only read? | Stage 1 candidate | Stage 2 or later |
| Could it start a job or spend capacity? | Pilot first | Continue |
| Could it write, delete or handle a credential? | Owner and review required | Continue |
| Do you have a rollback? | Proceed | Wait |
Rollback
Switching a flag off, or choosing Reset to Defaults, removes your override. Changes reach connected sessions in about a minute, though assistants that cached the tool list may need a restart. That makes rollback quick, which is the point of staging.
What gating does not cover
Tool gating controls Safeguard's own tools. It does not govern other MCP servers your agents use. For that you need a different control, see tool gating versus Guard.
Quick check
Can I enable tools per team?
Not through tool gating, which resolves per tenant. Use separate tenants or roles where the difference matters.
How do I undo a change?
Toggle the tool off, or use Reset to Defaults. Allow about a minute and restart cached sessions.
Sources
Last reviewed: 19 October 2026.
