Safeguard
Compliance•

What evidence supports AI assistant and MCP questionnaire answers?

A short evidence pack for AI assistant and MCP answers: policy, allowlist, register, flow diagram, sample logs and review records, each with a date.

Safeguard Security Team
3 min read

Prepare seven items: the AI use policy, the approved tools list, the MCP server register, a data flow diagram, a permission matrix for agents, a sample redacted log entry and records of your latest review. Date each one, name an owner, and keep them in one folder.

Why bring evidence to the first reply

Questionnaire threads grow when answers are claims without proof. Each follow-up costs days. Offering an artifact beside the claim answers the next question before it is asked. It also protects you, because it forces answers to match what you can show.

The seven items

ItemSupports the answer onOwner
AI use policyGovernance, acceptable useSecurity lead
Approved tools list with datesWhich assistants are allowedIT
MCP server registerInventory and change controlPlatform
Data flow diagramWhat data reaches which modelArchitecture
Agent permission matrixLeast privilege, approvalsPlatform
Redacted sample logAuditabilitySecurity
Review recordOngoing oversightSecurity lead

Make each item reviewer ready

Policy. One to three pages, dated, with an owner. Long policies that nobody follows invite questions.

Approved tools. Show tool, version family where relevant, approval date, and the account type used.

Register. Export with names, source, local or remote, tools and credential scope. Remove internal hostnames and anything a stranger should not have.

Diagram. One page. Arrows for data, boxes for providers, a label on what is retained.

Permission matrix. Rows for agents, columns for systems, cells for read, write or none. Reviewers read these in seconds.

Sample log. Real format, fake values. Make sure it shows user, tool, time and outcome.

Review record. A meeting note or ticket showing someone looked at the register and policy on a date.

Handle sensitive content safely

Share through a controlled channel, ideally under a mutual confidentiality agreement. Watermark with the recipient and date. Never include tokens, keys or customer names. The MCP security guidance warns that tokens leaked through logs can carry broad scopes, so check samples twice before they leave.

Keep the pack fresh

Set a calendar reminder for each owner. A policy dated eighteen months ago does not help, even if it is accurate. Align the cadence with other work, such as quarterly access reviews. See how to structure a questionnaire evidence folder and owners and review dates for answer libraries.

Where tooling helps

Safeguard's compliance and evidence automation collects records in one place, framework mapping shows which controls an artifact supports, and asset discovery feeds the inventory.

Try it yourself

Click through it at your own pace, or open it full screen. Look at how one piece of evidence can support more than one control.

Quick check

Should the pack be public?

Policy summaries and subprocessor lists can be public. Registers and logs should go out under confidentiality, per request.

How many items are too many?

If the owner cannot keep an item current, drop it. Seven fresh items beat fifteen stale ones.

Sources

Last reviewed: 19 October 2026.

Never miss an update

Weekly insights on software supply chain security, delivered to your inbox.

Self-healing security runs on Safeguard.

Your first fix PR is minutes away.

No sales call required, even your agent can complete the purchase over MCP.