Many people already spend their day inside Claude or ChatGPT. Asking them to open another app to check a package is a tax. Safeguard avoids it: you connect your assistant to Safeguard over MCP, and ask security questions where you already are.
What it is
MCP, the Model Context Protocol, lets an AI assistant call external tools. Safeguard runs an MCP server at mcp.safeguard.sh. Once your assistant is connected, it can look up your projects, findings and vulnerabilities.
It works with Claude, ChatGPT, Cursor or your own agent. Tool calls are unlimited on every plan.
You connect your existing Claude or ChatGPT account to Safeguard. Which model answers is your choice. Safeguard also lets you bring your own model, including Claude, GPT, Mistral and Cohere, and local or on premises models.
How to use it
The exact screens differ by assistant, and assistants change them often, so follow the current instructions for your client. The shape is always the same.
- In your assistant, add a new MCP server or connector.
- Use the address
https://mcp.safeguard.sh. - Sign in to Safeguard when asked, and approve access.
- Ask a question that needs real data.
Which of my projects have critical findings with a fix available?
What to check after connecting
- Ask something you already know the answer to, and confirm it matches the web app.
- Ask about a project outside your access, and confirm you do not see it. Access follows your Safeguard permissions.
- Ask the assistant to say which tool it called, so you know the answer came from Safeguard and not from memory.
Control for admins
Tenant admins can switch off any client type, including MCP clients, ChatGPT and Claude individually. If your policy is that AI assistants should not reach security data, that is one setting.
Try it yourself
The demo below walks through MCP server setup step by step. Do the exercise after it with a project you know well.
Click through it at your own pace, or open it full screen.
Quick check
What address do you add as the MCP server?
How do you confirm an answer came from Safeguard?
Ask the assistant which tool it called, and compare the result with the web app.
