Fix a vulnerable dependency with the Safeguard MCP server in ChatGPT

You can connect ChatGPT to the Safeguard MCP server at mcp.safeguard.sh, ask which dependency is vulnerable, and ask for a remediation plan for npm, pip, Maven, Go or Cargo projects. ChatGPT gives you the plan. You or the Safeguard CLI make the change in your repository, and you review it before merging.

This how-to follows Safeguard's documentation for the ChatGPT connection. Tool calls on the Safeguard MCP server are unlimited on every plan. What ChatGPT can do depends on which tools your administrator has enabled, so read the limits section before you start.

What do you need first?

• A Safeguard account at app.safeguard.sh with at least one project scanned.
• A ChatGPT workspace where custom MCP servers are allowed. The Safeguard docs say account and workspace policies apply, so if the option is missing, ask your ChatGPT workspace administrator.
• A Safeguard administrator, if you need tools beyond the default set.

How do you connect ChatGPT?

1. In ChatGPT, open Plugins, select the plus button, then choose Add custom MCP server.
2. Enter a name, for example Safeguard Security, and a description.
3. Under Connection, enter the server URL: https://mcp.safeguard.sh/mcp
4. Choose OAuth for authentication. ChatGPT registers itself, so there is no client ID or key to copy.
5. Read the risk warning, select I understand and want to continue, then select Create as a plugin.
6. When ChatGPT sends you to Safeguard, sign in, check the app name and the host that will receive the authorization code, then click Authorize access.

Nothing is installed on your machine. The docs say Safeguard publishes no npm package or local bridge for the MCP server, and any package claiming to be one should be treated as untrusted.

Open the MCP server setup demo in its own page.

How do you find the vulnerable dependency?

Ask in plain language. The docs list example prompts such as listing your security projects and finding critical vulnerabilities. A new workspace starts with six read-only tools, including listing projects, getting a project with its components and vulnerabilities, listing vulnerabilities, getting one CVE with its remediation, and listing findings. To check the connection, ask "What Safeguard tools do you have?" and look for tools such as safeguard_list_projects.

A good sequence:

1. "List my security projects."
2. "Find critical vulnerabilities in the payments project."
3. "Show details for CVE-2021-23337," or whichever CVE appears. The CVE tool returns the CVSS score and remediation.

How do you get the fix?

Ask: "Generate a remediation plan for my npm project." The docs say the server can generate remediation plans for npm, pip, Maven, Go and Cargo projects. A plan is advice: which package to move to which version, and why. It is not a merged change.

The remediation-plan tool may not be in the default six, so it can be disabled in your workspace. If ChatGPT reports that a tool is not enabled for the tenant, copy the tool name and ask a Safeguard administrator to enable it under Feature Flags. The docs say a change usually reaches a connected session within about a minute, but many clients cache the tool list, so reconnect if it does not appear.

How do you apply the change?

You have two safe options.

• Edit and test yourself. Bump the version in your manifest, reinstall, and run your tests.
• Use the Safeguard CLI. Its fix command updates vulnerable dependencies. Check the flags first:

safeguard fix --help

Either way, review the diff and run the test suite before merging. A version bump that fixes a CVE can still break behavior, which is why Safeguard's IDE remediation offers Safe, Balanced and Aggressive strategies. Start with the least risky.

What are the limits?

The default tools are read-only, so a fresh connection can answer questions but not change anything. ChatGPT can summarize a plan wrongly, so compare its words with the CVE details the tool returned. And a Custom GPT Action, the alternative path in the docs, uses one key for everyone who can use that GPT, so keep any such GPT private.

What to do

1. Create or confirm your Safeguard account and scan a project.
2. Add the custom MCP server in ChatGPT using the URL above and OAuth.
3. Ask for the tool list to confirm the connection.
4. List critical vulnerabilities and open one CVE.
5. Ask for a remediation plan, and ask an administrator to enable the tool if needed.
6. Make the change in your repo, run tests and review the diff before merging.

Quick check

Sources

• Safeguard docs, ChatGPT (OpenAI)
• Safeguard docs, MCP tool availability and gating
• Safeguard docs, MCP connector setup
• Safeguard docs, CLI usage
• Safeguard docs, IDE support

Last reviewed: 18 October 2026.