Add Safeguard to ChatGPT as a custom MCP server at https://mcp.safeguard.sh/mcp, sign in with OAuth, then ask for your critical vulnerabilities and a remediation plan in plain language. You only see what your Safeguard role and your workspace's enabled tools allow.
The steps below follow the Safeguard documentation for ChatGPT, read on 18 October 2026. Menu names in ChatGPT can differ by plan, so trust the docs over this summary if they disagree.
How do you connect it?
You need a Safeguard account at app.safeguard.sh first. Then, per the docs:
- In ChatGPT, open Plugins, select the plus button, then Add custom MCP server.
- Enter a name, for example Safeguard Security, and a description.
- Under Connection, enter https://mcp.safeguard.sh/mcp.
- Choose OAuth. ChatGPT registers itself with Safeguard, so there is no client id or key to copy.
- Review the risk warning, select I understand and want to continue, then Create as a plugin.
- When ChatGPT sends you to Safeguard, sign in, check the app name and the host the authorization code will go to, and click Authorize access.
If you do not see the option, ask your ChatGPT workspace administrator. Nothing is installed locally. The docs also warn that Safeguard publishes no npm package or local bridge for its MCP server, so treat any package claiming to be one as untrusted.
An alternative is a Custom GPT Action that imports https://mcp.safeguard.sh/openapi.json with an MCP key. That route has a smaller tool set and a shared key, so the docs advise keeping that GPT private.
What can ChatGPT see?
The assistant acts as you. The documentation says a new workspace starts with a small read-only set of tools, and an administrator turns on the rest per tool. That is the right default for triage, since reading comes before changing anything. Tool calls on the Safeguard MCP server are unlimited on every plan.
The listed capabilities include querying vulnerabilities, SBOMs, packages, licenses, findings and compliance data, generating remediation plans for npm, pip, Maven, Go and Cargo projects, comparing SBOMs, and checking policy gates.
What does a triage session look like?
Use the documented example prompts as a base and add context.
Start broad: "Find critical vulnerabilities." Read the list and note which projects own them.
Narrow it: ask which of those sit in projects that ship to production, and which packages are involved. If your workspace has the relevant tools enabled, ask for the licenses or SBOM entries that tie a package to a project.
Ask for a plan: "Generate a remediation plan for my npm project." The docs state this works for npm, pip, Maven, Go and Cargo. Read the plan as a proposal. Check the target version against the advisory before you merge anything.
Finish with a gate check, if enabled, to see whether the change would clear your policy. Record the outcome in your ticket.
We do not describe more than the docs state. If a step needs a tool your admin has not enabled, the assistant will not have it, and that is by design.
Where does this fit with the other surfaces?
The same findings appear in the web app, the CLI and the IDE extensions. MCP is a conversational layer over them. Use it for quick questions and planning. Use the CLI and CI for gates that must be repeatable.
What to do
- Confirm your account and role, and ask your admin which MCP tools are enabled.
- Add the server in ChatGPT with OAuth and check the authorization host before you click Authorize.
- Start with a read-only question about critical findings.
- Group results by project and owner before you pick what to fix first.
- Request a remediation plan for one project and verify versions against the advisory.
- Apply the change in a branch and let CI run your normal scan.
- Record what you decided and why in your tracker.
Try the walkthrough yourself at /demos/mcp-server-setup.
Quick check
Do I need to create an API key to use Safeguard in ChatGPT?
Not for the recommended route. OAuth handles sign in. Keys only apply to the Custom GPT Action alternative.
Can ChatGPT change things in my workspace?
Only through tools your administrator has enabled. A new workspace starts with a small read-only set.
Sources
- Safeguard MCP integration for ChatGPT, Safeguard Docs, read 18 October 2026
- MCP Integrations, Safeguard Docs, read 18 October 2026
Last reviewed: 18 October 2026.
