The Loop Returns MCP server lets you connect your AI assistant — like Claude, Cursor, or ChatGPT — directly to your Loop data. Once connected, your assistant can search returns, take actions, read policies, and manage workflows without leaving the conversation. This article covers how to connect, what tools are available, and what to expect from this early-access preview. Authentication uses OAuth 2.1 with PKCE — no API keys or secrets to manage.

Use the menu on the left to quickly navigate this article.

Your MCP server URL is:

https://api.loopreturns.com/mcp

In Claude Desktop or claude.ai, go to Settings → Connectors, click Add, and enter the server URL above. Claude handles the OAuth flow automatically — you'll be prompted to log in with your Loop admin credentials the first time.

If you add the server to your claude_desktop_config.json directly, use mcp-remote as a bridge (Claude does not yet support OAuth natively in that path):

{
  "mcpServers": {
    "loop-returns": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://api.loopreturns.com/mcp"]
    }
  }
}

Node.js is required. On first use, mcp-remote will open a browser window to complete the OAuth login.

Open your MCP configuration (Settings → MCP or your project's .cursor/mcp.json) and add:

{
  "mcpServers": {
    "loop-returns": {
      "url": "https://api.loopreturns.com/mcp"
    }
  }
}

Cursor will open a browser window for authentication when the server is first accessed.

If your client isn't listed, let us know — the server implements standard OAuth 2.1 discovery via /.well-known/oauth-authorization-server, so adding new clients is straightforward.

When you log in, you select which shop to connect to. Your session is scoped to that shop. To switch shops, log out of the MCP server and re-authenticate, selecting a different shop during login. We're aware this flow can be cumbersome for multi-shop merchants and are working on improvements.

The MCP server currently provides 22 tools across several categories.

Note: Workflow tools are exploratory. We're curious whether you find them useful and what other configuration or automation capabilities you'd want accessible through AI.

Some MCP clients — notably Claude — support Generative UI, which means the assistant can render interactive visual components alongside its text responses. When you use return management tools through a supported client, you'll see rich cards, tables, status badges, and action buttons rendered inline in the conversation.

[SCREENSHOT: List Returns with Generative UI — structured table with key return columns]

[SCREENSHOT: Return Detail Card — line items, resolution info, and status badges]

[SCREENSHOT: Post-Action Confirmation — updated status and quick-navigation buttons]

Note: Generative UI is purely additive. Every tool returns structured data regardless of whether the client can display visual components. Automated agents, scheduled tasks, and text-only assistants all receive the full data payload — nothing is lost without the UI layer.

If you run into any issues connecting, have ideas, or want to talk through use cases, please reach out to support@loopreturns.com.