MCP Server: make your store callable by any AI agent
The nexCommerce MCP server publishes your commerce capabilities — product search, availability, pricing, cart, checkout and order status — as Model Context Protocol tools. MCP is the open standard that lets AI agents call external systems safely. With it, an assistant can act on your store through governed, audited tools instead of scraping pages.
Key takeaways
- One MCP endpoint exposes the same catalogue, pricing, inventory and order services your storefront uses.
- Tools are scoped: read-only discovery for anonymous agents, transactional tools for authenticated sessions, B2B tools for contract buyers.
- Every call is authenticated, rate-limited, logged and attributable to a channel in the Commerce Console.
- The server is the foundation for agentic checkout protocols and for your own internal copilots.
What is an MCP server, in commerce terms?
An MCP server is a service that describes what an agent may do — search products, get availability, create a cart, get order status — and executes those actions when an agent asks. It replaces brittle page scraping with a contract: the agent sees typed tools, your platform sees an authenticated caller.
Which tools does nexCommerce expose?
Tools are grouped by what they touch and gated by who is calling. Discovery is public and rate-limited; anything that reads a customer or moves money needs an authenticated session.
| Tool group | Example tools | Access |
|---|---|---|
| Discovery | search_products, get_product, list_categories, get_availability_by_location | Public, rate-limited |
| Pricing & offers | get_price with customer context, get_applicable_promotions, get_delivery_slots | Public or session |
| Transact | create_cart, update_cart, apply_coupon, create_checkout, get_checkout_status | Authenticated session |
| After sale | get_order_status, get_shipment_eta, request_return | Authenticated customer |
| B2B | get_contract_price, create_quote, reorder_from_history | Authenticated buyer |
| Operations | catalogue_quality_report, forecast_sku, pick_priority | Staff roles via RBAC |
How does an ecommerce MCP server stay secure?
Scoped tokens per tool group, per-client rate limits, full request logging, and role-based access inherited from the Commerce Console. Prices and stock are computed server-side on every call, so an agent never receives more than the storefront would show the same customer.
- 01Register the agent client in the Commerce Console and choose its tool scopes.
- 02The agent discovers the available tools from the MCP endpoint.
- 03Each call is authorised, executed through the platform APIs and logged with channel attribution.
- 04Orders created by agents appear in the console with their source, agent identity and consent record.
How does MCP relate to ACP, UCP and AP2?
MCP is the tool layer. ACP, backed by OpenAI and Stripe, UCP, backed by Google and Shopify, and AP2, which covers payment mandates, are commerce and payment protocols that ride on top of it. The MCP server is the substrate for all three.
Agentic checkoutTalk to an engineer about the tool surface
Which tools to expose, to whom, and under what limits is an architecture conversation rather than a sales one.
Can I build my own internal agents on it?
Yes. The same server powers internal operations agents; expose additional tools under your own RBAC roles.
Does it expose customer data?
Only to authenticated sessions with the customer's consent, never to anonymous discovery tools.
Can I throttle or switch off a channel?
Per-client rate limits and kill switches are managed in the Commerce Console, so a misbehaving agent is contained without a deploy.