Gatlasdocs

Using MCP access

A guide for MCP recipients: how to connect your AI assistant and start querying data.

What you've been given

Someone in your organization has set up a data pipeline that syncs your business data (orders, customers, invoices, and more) into a centralized warehouse. They've given you an API key that lets your AI assistant query this data directly. You don't need a Gatlas account, just your API key and an AI tool that supports MCP.

Setup

1

claude.ai or Claude Desktop

Go to Settings → Connectors → "Add custom connector", paste this Server URL, and click Add:

https://www.gatlas.ai/api/mcp

Click "Connect" and a Gatlas authorization page opens. Paste your API key there and click Authorize. Nothing is typed into Claude directly; Claude never sees the raw key.

2

Claude Code (terminal)

Run this command in your terminal, replacing YOUR_API_KEY with the key you received:

claude mcp add --transport http gatlas https://www.gatlas.ai/api/mcp --header "Authorization: Bearer YOUR_API_KEY"
3

Cursor and other MCP clients

Add this to your client's MCP configuration (e.g. ~/.cursor/mcp.json):

{
  "mcpServers": {
    "gatlas": {
      "url": "https://www.gatlas.ai/api/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_API_KEY"
      }
    }
  }
}

Note

After adding the server, restart your AI tool. You should see Gatlas listed as a connected MCP server with available tools.

What you can ask

Once connected, you can ask your AI assistant questions in plain language. Here are some examples:

Querying data

  • "What were total revenues last quarter?"
  • "Show me the top 10 customers by order value"
  • "Compare monthly recurring revenue across the last 6 months"

Exploring available data

  • "What data sources are available?"
  • "What tables do I have access to?"
  • "Describe the monthly_revenue view"

Checking health

  • "Are all data syncs running successfully?"
  • "When was the last sync for Shopify?"
  • "Which connections have errors?"

Troubleshooting

  • "Tool not found": make sure you've added the MCP server correctly. Check that the URL is https://www.gatlas.ai/api/mcp (with the www.; the bare gatlas.ai form can fail sign-in in some clients) and that your API key reached Gatlas, via the Authorization header for Claude Code and Cursor, or via the Gatlas authorization page for claude.ai and Claude Desktop.
  • "Unauthorized" or "Invalid API key": double-check your API key. It starts with gat_. If it's not working, ask your admin to generate a new one.
  • "No data returned": your access may be scoped to specific entities or views. Ask your admin what data you have access to.
  • "View not found": the view may not be published to MCP, or it may not be in your scope. Ask your admin to check that the view has MCP visible enabled and is tagged to your entities.