Sign in
Guide / MCP

MCP for AI agents

Connect Claude Desktop, Cursor, Cline, or any MCP-compatible client to Giftronaut. Send gift cards, manage recipients, and check balance using natural language — no SDK installation or custom integration code required.

This documentation is in beta and is being updated regularly. If you have any questions or feedback, please contact us at api@giftronaut.com.
One credential, two ways to connect. Giftronaut MCP uses the same API credentials (client_id and client_secret) as the REST API. You do not need a separate account, credential set, or billing setup. The difference is authentication: MCP requires an interactive, browser-based sign-in. It cannot use server-to-server access tokens. MCP activity also appears in the same request log dashboard you use for REST API requests.

What is MCP?

The Model Context Protocol (MCP) is an open standard from Anthropic for connecting LLM agents to external tools and data. A Giftronaut MCP server hosts a typed catalog of 47 tools across orders, recipients, balance, catalog, campaigns, and templates — discoverable, schema-validated, and safety-guarded.

Compared to calling our REST API directly:

  • No setup code — add the Giftronaut MCP server URL. Your AI assistant automatically discovers all available tools and their parameters.
  • Built-in safety — for tools that can move funds or change data, Giftronaut adds safeguards such as confirmation checks, idempotency keys, daily USD spend limits, and per-tool controls.
  • Simple authentication — users approve access through a browser sign-in flow. No API keys to copy and paste.
  • Same observability — every tool call writes an audit_events row alongside your REST traffic, viewable in the portal with a single filter toggle.

Quick connect

Choose your MCP client and add the matching snippet to its MCP configuration file. The first tool call triggers the browser sign-in.

💡 Using an AI agent? You don't need to copy the snippet below manually. Paste the install prompt from MCP Client Setup into your assistant and it will configure itself automatically.

Claude Desktop

JSON · claude_desktop_config.json
{
  "mcpServers": {
    "giftronaut": {
      "url": "https://api.giftronaut.com/api/v1/mcp",
      "transport": "streamable-http"
    }
  }
}

Cursor

JSON · ~/.cursor/mcp.json
{
  "mcpServers": {
    "giftronaut": {
      "url": "https://api.giftronaut.com/api/v1/mcp"
    }
  }
}

OpenCode / Cline

JSON · ~/.config/opencode/opencode.json
{
  "mcp": {
    "giftronaut": {
      "type": "remote",
      "url": "https://api.giftronaut.com/api/v1/mcp"
    }
  }
}
Do not add an Authorization header. The SDK negotiates OAuth automatically — adding a static bearer token can force the client into legacy authentication mode and skip the user consent flow. See MCP Client Setup for the full setup guide.

Your first prompt

Once connected, just talk to your assistant. Try:

"Send a $25 Amazon gift card to john@example.com with the message 'Welcome to the team!'"

Behind the scenes, the agent will:

  1. Call giftronaut_search_branded_cards to locate Amazon in the catalog
  2. Call giftronaut_draft_order to create a preview without moving funds (1-hour token lifetime)
  3. Show you the draft summary and ask for confirmation
  4. On your approval, call giftronaut_execute_order to place the real order — the response carries the created orderId
MCP is production-only. Sandbox credentials cannot be selected on the MCP sign-in screen and existing sandbox tokens are rejected with 401 invalid_token (they still work for the REST API). Real gift cards are issued over MCP — for a safe first test, preview the email with giftronaut_send_test_email, use dryRun=true on cancel/resend, and keep the draft → execute confirmation flow.

Next steps