Can ChatGPT gate a pull request through the Safeguard MCP server?

Partly. ChatGPT can connect to Safeguard's MCP server, read your findings and ask for policy gate checks, so it can help you decide whether a pull request is safe to merge. The PR Guard review itself runs inside Safeguard and posts to GitHub. ChatGPT is a reader and an advisor, not the merge button.

What does the documentation say ChatGPT can do?

Safeguard's ChatGPT page says that once connected, ChatGPT can query vulnerabilities, SBOMs, packages, licenses, findings and compliance data, generate remediation plans for npm, pip, Maven, Go and Cargo projects, compare SBOMs, and enforce policy gate checks for deployment readiness. Those are the verified abilities. The page does not list a tool that runs a PR Guard review, so do not plan around one.

What does PR Guard do?

PR Guard is the AI review of a pull request's diff. In Safeguard you open a project's PR Guard tab, pick a pull request and click Run Review. It returns a score, a summary and comments tied to files and lines, each with a severity from critical down to info, and sometimes a suggested fix. Tick Post to GitHub before running and the comments also appear on the pull request. Today it works with GitHub repositories only.

So a realistic split is: PR Guard reviews the diff, and ChatGPT helps you read the surrounding picture.

How do you connect ChatGPT?

Safeguard hosts the server, so nothing is installed. The documented steps:

1. Create or sign in to your account at app.safeguard.sh.
2. In ChatGPT, open Plugins, select the plus button, then Add custom MCP server.
3. Give it a name, such as Safeguard Security, and a description.
4. Under Connection, enter the server URL https://mcp.safeguard.sh/mcp.
5. Choose OAuth. ChatGPT registers itself, so there is no key to copy. Review the risk warning, select I understand and want to continue, then 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, and click Authorize access.

Whether you see the option depends on your ChatGPT account and workspace policy. If it is missing, ask your workspace administrator. Do not install any package that claims to be the Safeguard MCP server, because Safeguard publishes none.

Why can ChatGPT see so few tools at first?

By design. A new workspace starts with six read-only tools: listing products and projects, getting a project, listing and getting vulnerabilities, and listing findings. Everything else, including tools that start scans or add integrations, is off until a tenant administrator enables it under Feature Flags. If an assistant calls a tool that is off, the server returns a normal result with an error flag saying it is not enabled for the tenant.

This suits gating work. Reading findings is low risk. Starting scans is a deliberate choice an administrator should make.

What would a review conversation look like?

Keep prompts specific. These are examples of the kind of request the docs describe, not guaranteed outputs:

• List my security projects, then open the project for this repository.
• Show critical and high findings for it.
• Which of those are new compared with the last SBOM?
• Do the policy gate checks pass for this project?

Use the answers as input to a human decision. Check the project in Safeguard before trusting a summary, because a chat model can misread or skip rows.

What are the limits?

Merging is yours. Nothing here says ChatGPT blocks or approves a merge on GitHub. Branch protection and required reviews stay the real gate.

Prompt injection. Text in a pull request, such as a comment or a file, can contain instructions aimed at the model. Keep ChatGPT's access read focused until you trust the setup.

Plan terms. What ChatGPT allows for custom MCP servers on your plan is OpenAI's decision. Safeguard says tool calls through its server are unlimited and never metered.

Shared keys. The Custom GPT Action alternative uses one key for everyone who can use that GPT, and the key acts as you. Keep such a GPT private.

What to do

1. Ask your administrator whether ChatGPT is allowed as a client for your tenant.
2. Connect with the OAuth steps above.
3. Start with the six default read tools.
4. Run PR Guard in Safeguard for the pull request, with Post to GitHub if you want inline comments.
5. Ask ChatGPT for findings and gate status on the same project.
6. Compare both views, then decide as a person.
7. Keep branch protection on GitHub as the enforced gate.

Quick check

Sources

• Safeguard docs: ChatGPT (OpenAI)
• Safeguard docs: Tool availability and gating
• Safeguard docs: PR Guard

Last reviewed: 18 October 2026.