Skip to content

Govern your agents in 5 minutes

You'll connect an AI assistant to WhiteBoxXAI over the MCP server, tell it the rules it must follow, and check that its first answer and its first record come from your data. For the longer version, with compliance questions and a governance review, see Connect an AI assistant over MCP.

You need: an MCP-capable client (Claude Desktop, Claude Code, Cursor or VS Code), Python 3.11 or later, and a WhiteBoxXAI API key (administrators create these in Settings → API keys).

1. Install

pip install whiteboxxai-mcp

2. Connect your client

For Claude Desktop (claude_desktop_config.json) or Cursor (~/.cursor/mcp.json), add:

{
  "mcpServers": {
    "whiteboxxai": {
      "command": "whiteboxxai-mcp",
      "env": {
        "WHITEBOXXAI_MCP_API_BASE_URL": "https://api.whiteboxxai.com",
        "WHITEBOXXAI_MCP_API_KEY": "wbx_live_..."
      }
    }
  }
}

For Claude Code:

claude mcp add whiteboxxai \
  --env WHITEBOXXAI_MCP_API_KEY=wbx_live_... \
  -- whiteboxxai-mcp

WHITEBOXXAI_MCP_API_BASE_URL defaults to https://api.whiteboxxai.com. Business Cloud tenants set it to the API address they were given. Restart the client so it starts the server. Other clients are on the MCP Server page.

3. Give the assistant its rules

Tell the assistant:

Read https://whiteboxxai.com/.well-known/skills/whitebox-governance-rules/SKILL.md and
https://whiteboxxai.com/.well-known/skills/whitebox-mcp-server/SKILL.md and follow them.

The rules tell it to report empty or unassessed results as they are, never invent a number, and never cast a governance vote. See Agent Skills.

4. Check that it works

Ask:

Which AI systems do we govern?

The assistant should call whitebox_models_list and answer with the names, versions and statuses it returned. If your account has no systems yet, a correct answer says so; it doesn't make some up. Load sample data if you want something to look at.

Then have it leave a record:

Has the first model on that list drifted? Create a drift report so we have a record, and give me its id.

The assistant should call whitebox_drift_detect, then whitebox_drift_create_report, and quote the report id. Find that report in the dashboard to confirm it exists: that's your check that the assistant reported what it did rather than what it expected to do.

What you've set up

  • An assistant that reads and records through the MCP Governance Layer, on your API key
  • Rules that keep it from inventing numbers or voting on your behalf
  • A drift report you can open in the dashboard and show to a reviewer

Revoke the API key in Settings → API keys when you're done experimenting. To go further, follow the 20-minute tutorial: compliance questions, alert triage, and preparing a review that a person decides.