Overview
After being granted access to a vendor's Trust Center, users can connect AI tool tools (Claude, ChatGPT, Cursor, or VS Code), directly to it using MCP (Model Context Protocol).
Once connected, users can utilize AI tools to ask questions about the vendor's security posture and have it search, browse, and pull documents, instead of manually browsing or clicking through the Trust Center.
This guide covers what the connection gives users, how to set it up, and what to do if something doesn't work.
What this enables
Once connected, AI tools can:
List the Trust Center users currently have access to
Browse a Trust Center's structure by product, page, and section
Search across certifications, reports, policies, questionnaires, and published updates
Get a time-limited download link for a specific document
All of this reflects exactly what users can already see in the Trust Center's web portal. Connecting AI tools does not grant any new access; it only gives the AI tool a way to read what the user is already permitted to see.
Before beginning
Important: Users need an active, approved access to the Trust Center before connecting. If the user has not already requested access and clicked the confirmation link sent to their email, they must do that first, then come back to these steps.
The Trust Center MCP URL is: https://app.safebase.io/api/trust-center/mcp
Which vendor the user is connected to isn't determined by this URL. It's determined by which Trust Center they're currently signed in to when the connection is approved.
If users have access to more than one vendor's Trust Center, see the FAQ below before connecting.
Step 1: Add the connector an AI tool
Every client below is doing the same thing: adding a new remote MCP server and pointing it at the URL above.
Users do not need a token, API key, or password at this step; authorization happens in step 2.
Claude (claude.ai) and Claude Desktop
Team/Enterprise: if custom connectors aren't already available, a Claude Workspace owner or admin must first enable them under Organization settings > Connectors > Add > Custom > Web.
Once enabled, anyone on the workspace can connect using the steps below.
Open Customize > Connectors.
Click the + button, then Add custom connector.
Enter the Trust Center's MCP URL.
Sign in to the Trust Center when prompted.
To use it in a chat, click the + button in the lower left, select Connectors, and enable it.
ChatGPT (Plus, Pro, Business, Enterprise, or Edu)
Custom MCP connectors require Developer mode. For Team, Enterprise, and Edu workspaces, a workspace admin must first enable it under Admin Workspace Settings > Permissions & Roles > Connected Data > Developer mode.
Open Settings > Connectors.
Scroll to Advanced and toggle Developer mode on.
Click Create.
Enter the Trust Center's MCP URL.
Sign in to the Trust Center when prompted.
Claude Code
Run:
claude mcp add --transport http trust-center https://app.safebase.io/api/trust-center/mcp
Inside Claude Code, run /mcp to complete sign-in in browser. See Connect Claude Code to tools via MCP
Cursor
Add the server to ~/.cursor/mcp.json (or .cursor/mcp.json in a project):
{
"mcpServers": {
"trust-center": {
"url": "https://app.safebase.io/api/trust-center/mcp"
}
}
}
Restart Cursor and sign in when prompted. See Model Context Protocol (MCP)
Codex CLI
Run:
codex mcp add trust-center --url https://app.safebase.io/api/trust-center/mcp
First time use requires sign. See MCP in Codex
Windsurf
Add the server to ~/.codeium/windsurf/mcp_config.json:
{
"mcpServers": {
"trust-center": {
"serverUrl": "https://app.safebase.io/api/trust-center/mcp"
}
}
}
Sign in when prompted. See Cascade MCP Integration
VS Code
Run MCP: Add Server from the Command Palette, or add the server directly to .vscode/mcp.json:
{
"servers": {
"trust-center": {
"type": "http",
"url": "https://app.safebase.io/api/trust-center/mcp"
}
}
}
Sign in when prompted. See Use MCP servers in VS Code.
Step 2: Authorize access
When users first use the connection, the AI tool opens a browser window to a SafeBase | Authorize access screen. What users see depends on the session:
If already signed in to that vendor's Trust Center: users see "Continue as [email] for [Trust Center name]."
Click it to approve the connection.
Click Cancel to back out instead.
If not signed in to any vendor Trust Center: the screen tells users to request access to the Trust Center and click the link in the confirmation email, then try connecting again.
Important: Only approve a connection if you started the request yourself and recognize the application and destination shown on the screen.
Once users approve, the AI tool is connected to that Trust Center. Users won't need to repeat this step for that tool and vendor unless access is revoked or the session expires.
What can asked
Users don't need to know any tool names or commands. Ask the AI tool questions the way someone would ask a person, for example:
"What certifications does this vendor have?"
"Show me their subprocessors."
"Do they have a recent penetration test report I can download?"
"What Trust Centers do I currently have access to?"
"Has anything changed with their compliance status recently?"
The AI tool decides which underlying action to take based on the question.
Search returns quick, specific facts; browsing the Trust Center's structure surfaces the fuller picture when a search comes back empty or unclear.
Tools available
Below is a list of tools available in the Trust Center MCP for reference to see exactly what's exposed. Every tool is read-only and none of them can change anything in the vendor's trust center.
Tool | What it does |
tc_access | Lists every Trust Center users have access to, with each vendor's organization name and public URL. Flags the Trust Center the current session belongs to. |
tc_explore | Navigates a Trust Center's structure, products, pages, and content groups such as certifications, reports, policies, and questionnaires, drilling from a high-level overview down to full item detail. |
tc_search | Searches across a Trust Center's documents, knowledge base entries, content items, and published updates for a specific fact. Can also browse everything of a given content type. |
tc_document_download | Returns a time-limited, signed download URL for a document, using a document ID surfaced by tc_search or tc_explore. |
Frequently asked questions
Do I need a password or API key to connect?
No. Authorization happens through the same email-based access flow you already use to view the vendor's Trust Center. There's no separate credential to create or store.
What can my AI tool access?
Access is scoped to your account and permission level, exactly as it is when you browse the Trust Center directly.
I have access to more than one vendor's Trust Center. Do I need to connect separately for each?
Yes, each vendor is a separate connection with its own URL. Ask your AI tool to list your accessible Trust Centers to confirm which ones you're connected to.
Do document download links expire?
Yes. Links your AI tool retrieves are time-limited. If a link stops working, ask your AI tool to fetch a new one.
How do I disconnect?
Remove the connector or server entry from your AI tool's settings or config file. This only removes the connection; your underlying access to the vendor's Trust Center is unaffected.