Connect ChatGPT to Safeguard by adding https://mcp.safeguard.sh/mcp as a custom MCP server with OAuth. Scan timing is not set from the chat. Frequency is configured in Scan Policies in the Safeguard platform, and ChatGPT then asks about the latest results in plain language. Tool calls on the MCP server are unlimited on every plan.
What can the connection do, and what can it not?
The Safeguard MCP server lets an AI assistant query your software supply chain data, such as vulnerabilities, SBOMs, findings and compliance status. It also works with Claude, Cursor or your own agent.
One boundary is worth stating plainly. The documentation we checked does not describe creating a scan schedule through ChatGPT. Scheduling lives in the platform. So the useful pattern is two parts: set the schedule once in Safeguard, then use ChatGPT to read and explain the results.
How do you add it to ChatGPT?
These are the steps from the Safeguard setup page.
- In ChatGPT, open Plugins, select the plus button, then Add custom MCP server.
- Enter a name and a description.
- Under Connection, paste
https://mcp.safeguard.sh/mcp. - Set authentication to OAuth, read the risk warning, accept it, and select Create as a plugin.
- Sign in to Safeguard and click Authorize access.
Whether you can add a custom MCP server depends on your ChatGPT account and workspace policy. If the option is missing, ask your workspace admin. Safeguard tenant admins can also switch off any client type, including ChatGPT, so a blocked connection may be a deliberate setting on that side.
How does Safeguard decide when to rescan?
The continuous scanning documentation describes three kinds of trigger: updates to vulnerability feeds, such as new CVEs, GHSA and OSV entries, CISA KEV additions or EPSS changes; Safeguard zero day advisories; and asset changes such as a new commit, a new image tag or a vendor SBOM upload.
Default schedules by asset, as documented:
- Source repositories: on every push, plus an hourly pull in case a webhook was missed.
- Container images: on push, and every 4 hours for running images.
- Kubernetes workloads: on admission, plus a runtime reconciliation every 15 minutes.
- AI models: on publish, plus weekly revalidation.
- SBOMs: on upload, plus hourly feed driven rescoring.
You can tune frequency under Policies, then Scan Policies. The docs give an example of a 5 minute interval for workloads labeled as sensitive production and 24 hours for development or sandbox environments. Alert routing is under Settings, then Notifications, and Slack, Teams and Discord channels can receive alerts.
What would you ask ChatGPT once it is connected?
Keep questions concrete and tied to your data:
- Which critical findings were raised since Monday, and which have a fix available?
- Which of my repositories have not been scanned in the last day?
- Summarize the findings on the payments service for an engineering lead.
Check the first few answers against the web app until you trust the connection. An assistant summarizes what the server returns, and you remain responsible for the decision.
What are the risks to manage?
Any connection from a chat client to a security data source widens access. Use a role with the narrowest rights that still answers your questions. Review who in your workspace can use the plugin. Remember that findings can be sensitive, so apply the same rules you would to a dashboard export.
For on demand scans, the CLI documents scanning a single asset by its ID and scanning every asset in a project. See the CLI help for the exact flags. The docs also show a webhook example, but note that the /v1 REST endpoints are not yet available, so check the API reference for what works today.
What to do
- Confirm your ChatGPT workspace allows custom MCP servers.
- Add the server using the steps above and authorize with a least privilege Safeguard role.
- Open Policies, then Scan Policies, and set intervals by asset label.
- Configure alerts under Settings, then Notifications.
- Ask three test questions and verify the answers in the web app.
- Write down which actions stay manual, such as triggering a scan through the CLI.
- Review plugin access each quarter.
Quick check
Can ChatGPT create or change a scan schedule?
The documentation we reviewed does not describe that. Set schedules in Scan Policies and use ChatGPT to query results.
Does the MCP server limit how many tool calls I make?
No. Tool calls are unlimited on every plan. Your ChatGPT plan has its own limits, which are separate.
Sources
- Safeguard MCP server setup
- Safeguard documentation: continuous scanning
- Safeguard documentation index
Last reviewed: 18 October 2026.
