Can you prepare audit evidence with the Safeguard MCP server in ChatGPT?

Partly. You can connect the Safeguard MCP server to ChatGPT with OAuth and ask it to list projects, vulnerabilities and findings, which are useful audit inputs. Framework evidence reports are built in the Safeguard web app under Reports. Treat ChatGPT as a fast way to query the data, not as the audit record.

What can ChatGPT see through the server?

The server lives at mcp.safeguard.sh and is hosted, so nothing is installed locally. Safeguard's docs warn that it publishes no npm package or local bridge, so any package claiming to be the Safeguard MCP server should be treated as untrusted.

A new workspace starts with six read only tools: list products, list projects, get project, list vulnerabilities, get vulnerability and list findings. A workspace administrator can enable other tools one at a time under Feature Flags, each with its own flag. The docs do not name dedicated evidence or audit tools for MCP, and they do not document scheduled scans, so this post does not claim either. Ask your admin what is enabled in your tenant.

How do you connect ChatGPT?

Per the Safeguard docs:

1. Sign in to your account at app.safeguard.sh.
2. In ChatGPT, open Plugins, select the plus button, then Add custom MCP server.
3. Enter a name and a description.
4. Under Connection, enter the server URL https://mcp.safeguard.sh/mcp.
5. Set authentication to OAuth, read the risk warning, select I understand and want to continue, then Create as a plugin.
6. Sign in on the Safeguard page, check the app name and host, and click Authorize access.

If the option is missing, your ChatGPT workspace administrator controls custom MCP servers, and the docs say account and workspace policies apply.

Where does audit evidence really come from?

Safeguard's compliance docs describe mapping automated tests from connected systems to frameworks including SOC 2, ISO/IEC 27001, the HIPAA Security Rule, PCI DSS and NIST CSF 2.0. Each framework page separates requirements a scan decides from those that need evidence a scan cannot see, such as an access review or a policy document. For reports, the web app flow is Reports, Create Report, pick a framework, set scope and date range, then generate and export.

That means a sensible split:

• Use the web app to produce the report an assessor will receive.
• Use ChatGPT to answer questions while you prepare, for example which projects have open critical findings, or what a given vulnerability affects.

Why keep ChatGPT out of the evidence chain?

Assessors want records they can trace: who ran what, when, against which scope. A chat answer is a summary produced on demand. It can be useful to find the gap, but the evidence should be the exported report or the underlying finding, with its timestamp. An MCP key or OAuth session acts as the member who authorized it, so scope that member's access to what the audit needs.

Safeguard does not certify anyone against a framework. These tools help you collect and organize evidence for your own audit.

What to do

1. Ask your administrator which MCP tools are enabled, and enable only the read tools you need.
2. Connect ChatGPT using the steps above.
3. Prompt for a list of projects and their open findings, then compare it with the web app.
4. Build the actual report in the web app with the right framework, scope and date range.
5. Store the exported file with a date, and note who generated it.
6. Gather the non-scan evidence (policies, access reviews) separately.

Quick check

Sources

• Safeguard docs: MCP
• Safeguard docs: ChatGPT setup
• Safeguard docs: tool gating
• Safeguard docs: compliance

Last reviewed: 18 October 2026.