What is an MCP commerce server?
The Model Context Protocol (MCP) is a standard way to expose tools to AI agents like Claude and GPT. An MCP commerce server exposes the tools an agent needs to transact — catalog, checkout, and orders — so the model can place an order without you hand-wiring a tool schema per provider.
The problem MCP solves
Without a shared protocol, every model wants tools defined its own way, and every integration re-implements auth, schemas, and transport. MCP standardizes that: a server advertises tools and their input schemas, and any MCP-capable client (Claude Desktop, IDEs, agent runtimes) can call them.
What a commerce MCP server exposes
For an agent to buy something, the useful tools are:
create_checkout_session— open a checkout for catalog line itemsupdate_checkout_session— attach buyer and fulfillment detailscreate_payment_intent— a payment intent for the sessioncomplete_checkout_session— create an order, idempotent per session
Run OrderCore's MCP server
OrderCore ships ordercore-mcp, exposing 10 commerce tools (catalog search, product,
prices, inventory, order history, order trace, and the idempotent checkout flow). It is published to
npm and listed in the official MCP Registry
as ai.ordercore/ordercore-mcp.
Fastest path — no install, no signup. Run it with no API key and it issues a read-only sandbox key for the demo catalog automatically:
npx @ordercore/mcp
Standalone static binaries are also available from /downloads (macOS arm64 · macOS x64 · Linux · Windows), unzip, then register it with your client.
Claude Code (one command):
claude mcp add ordercore -- npx -y @ordercore/mcp
That runs read-only against the demo catalog with no key. Add your key to enable checkout:
claude mcp add ordercore --env ORDERCORE_API_KEY=oc_live_xxx -- npx -y @ordercore/mcp
Claude Desktop (claude_desktop_config.json) or Cursor (.cursor/mcp.json):
{
"mcpServers": {
"ordercore": {
"command": "npx",
"args": ["-y", "@ordercore/mcp"],
"env": { "ORDERCORE_API_KEY": "oc_live_xxx" }
}
}
}No API key yet? Get one in minutes at /bootstrap — no payment required.
Prefer raw HTTP? The same tools are plain REST — the full API is described by the OpenAPI 3.1 spec, and ready-made tool schemas for OpenAI, Claude, Gemini, DeepSeek, and Grok are served at /direct-ai/tooling. Full setup: Claude / MCP quickstart.
MCP vs raw REST — which to use
- MCP when your client already speaks it (Claude Desktop, agent runtimes) — zero per-model schema work.
- REST when you're wiring tools yourself or calling from a backend — full control, same endpoints.
Try it (no signup)
See the tools in action — the demo runs the checkout flow and proves idempotency offline in mock mode:
Related
- MCP for merchants — let AI agents shop and check out on your store
- How to give your AI agent a checkout tool
- Idempotent orders for AI agents
- OrderCore vs Stripe + a custom order backend
- Full API docs