How to Connect Claude to Your GEO Data: Refine MCP Setup for Claude Desktop, Claude Code & the API

Summary

Claude is a first-class MCP client: Claude Code connects to Refine with a single terminal command, Claude Desktop with a short JSON config, and the Claude API through its MCP connector. Once connected, Claude can pull your mention rate and visibility history, inspect the raw AI answers behind your scores, manage your monitored prompts and competitors, and generate publish-ready content. This guide walks through all three setups with copy-paste snippets.

What You Need Before Starting

• A Refine account on the Visibility or Autopilot plan, with at least one brand configured.
• A Refine API key — created from your dashboard, it starts with rfn_live_ and is shown only once. Store it in a password manager.
• The endpoint URL: https://app.refine-app.com/api/mcp/mcp — a remote Streamable HTTP MCP server, so there is nothing to install or host.

Connect Claude Code (One Command)

Claude Code registers remote MCP servers from the terminal. Run this once, replacing the key with your own:

claude mcp add --transport http refine \
  https://app.refine-app.com/api/mcp/mcp \
  --header "Authorization: Bearer rfn_live_your_key"

Restart Claude Code and the Refine tools appear automatically. Run /mcp inside a session to confirm the server is connected and see the fifteen available tools.

Connect Claude Desktop

Claude Desktop reads its MCP servers from claude_desktop_config.json (Settings → Developer → Edit Config). Add the refine block under mcpServers:

{
  "mcpServers": {
    "refine": {
      "type": "http",
      "url": "https://app.refine-app.com/api/mcp/mcp",
      "headers": {
        "Authorization": "Bearer rfn_live_your_key"
      }
    }
  }
}

Restart the app. You will see Refine listed among your tools, and Claude will ask for approval the first time it calls each tool — approve once and it remembers.

Connect via the Claude API

For programmatic agents, the Claude API ships an MCP connector: pass the server in the mcp_servers parameter of a Messages request and Claude calls the tools mid-conversation without you writing any tool-handling code.

response = client.beta.messages.create(
    model="claude-sonnet-5",
    max_tokens=2048,
    betas=["mcp-client-2025-04-04"],
    mcp_servers=[{
        "type": "url",
        "url": "https://app.refine-app.com/api/mcp/mcp",
        "name": "refine",
        "authorization_token": "rfn_live_your_key",
    }],
    messages=[{
        "role": "user",
        "content": "How is my AI visibility trending over 30 days?",
    }],
)

Prompts to Try First

• "List my monitored prompts and flag any that overlap or target the same question."
• "What is my mention rate and average rank over the last 30 days? How does the trend compare to the last 90?"
• "Pull the raw test results for my worst-performing prompt and tell me which competitors the AI engines recommend instead."
• "Generate a blog article for that prompt, grounded in these two source URLs, and show me the draft when it is ready."
• "Add [competitor] to my tracked competitors and tag all my pricing-related prompts with a new Pricing tag."

Troubleshooting

• 401 Unauthorized — the Authorization header is missing or the key is wrong. Check for a missing "Bearer " prefix, and confirm the key was not revoked in your dashboard.
• Tools not appearing — restart the client after editing config; in Claude Code, run /mcp to inspect connection status.
• Content generation returns a contentId but no text — generation is asynchronous by design. Ask Claude to poll get_content until status is completed; long articles can take a few minutes.
• Rate limited — each key allows 60 requests per minute. Agents in tight loops can hit this; tell the agent to batch reads instead of polling aggressively.