CartAgent

Agent Connect — MCP integration for AI agents

CartAgent exposes a public Model Context Protocol (MCP) server. Any agent that speaks MCP can search, browse, cart, and checkout against a live WooCommerce store. No SDK, no API key, no setup beyond a single config block.

What is the Model Context Protocol?

MCP is an open protocol from Anthropic for connecting AI assistants to external tools and data. It's supported natively by Claude Code, the Anthropic Agent SDK, OpenClaw, Claude Desktop, and a growing set of third-party clients including Cursor and Continue. CartAgent runs an MCP server at https://api.cartagent.store/mcp.

Connect in 30 seconds

Claude Code

Add to ~/.config/claude-code/config.json:

{
  "mcpServers": {
    "cartagent-demo": {
      "url": "https://api.cartagent.store/mcp"
    }
  }
}

Restart Claude Code, then prompt: "Use the cartagent-demo MCP to find me a network router and tell me the price."

OpenClaw

Add to your openclaw.json:

{
  "plugins": {
    "entries": {
      "cartagent-demo": {
        "type": "mcp",
        "transport": { "type": "http", "url": "https://api.cartagent.store/mcp" }
      }
    }
  }
}

Anthropic Agent SDK (Python)

from anthropic import Anthropic

client = Anthropic()
response = client.messages.create(
    model="claude-sonnet-4-6",
    max_tokens=1024,
    mcp_servers=[{"url": "https://api.cartagent.store/mcp", "name": "cartagent-demo"}],
    messages=[{"role": "user", "content": "Find me a WiFi 7 router in the cartagent-demo store"}]
)

Direct JSON-RPC (any HTTP client)

curl -X POST https://api.cartagent.store/mcp \
  -H "Content-Type: application/json" \
  -d '{
    "jsonrpc": "2.0",
    "id": 1,
    "method": "tools/call",
    "params": {
      "name": "search_products",
      "arguments": {"query": "wifi router", "limit": 5}
    }
  }'

The tool catalogue

ToolInputsReturns
search_productsquery, limit?, category?, min_price?, max_price?, in_stock?Up to 20 matching products with name, price, image, stock, URL
get_productproduct_idFull details: variants, attributes, images, description, stock
add_to_cartproduct_id, quantity, cart_token?, idempotency_keyUpdated cart_token + cart contents + totals
get_cartcart_tokenCart state — items, quantities, subtotal, tax, total
checkoutcart_token, customer details, payment_method (stripe / x402)Stripe checkout URL or USDC payment params
get_order_statusorder_idStatus, payment, shipping, tracking
Demo store is sandboxed. Checkout returns a working Stripe test-mode URL — but no real money moves and no real order is fulfilled. The 15 products are fictional. Use the demo to validate your agent's flow before pointing it at a production store.

Example agent prompts

Once connected, try these natural-language queries with your agent:

PromptWhat the agent does
"Find me a tri-band WiFi 7 router under $1500"Calls search_products({query: "tri-band WiFi 7", max_price: 1500}), summarises results
"Add the GL.iNet router to a cart and show me the total"Calls add_to_cart then get_cart
"What's in my cart right now? Tell me if any items are out of stock."Calls get_cart, cross-references stock via get_product
"Compare the GL.iNet router and the Roccat mouse on price and ratings"Two get_product calls, agent does the comparison itself
"Place an order for one of each Roccat mouse and Bluetooth speaker, payment via Stripe test mode"Full cart-build + checkout flow, returns Stripe URL

Rate limits

Public demo MCP is rate-limited per client IP. Limits are generous for normal agent use (~60 tool calls per minute). For higher limits or production access, register a store at cartagent.store/pricing and receive a token-gated MCP endpoint with your own quota + audit log.

Production usage

The demo MCP is a sandboxed copy of the same code that powers production CartAgent stores. When you register your own WooCommerce store via the WordPress plugin, you get:

→ See pricing  ·  → Try the chat bubble version first