SAP Commerce Cloud MCP Server and Agentic AI: What It Means for E-Commerce

At NRF 2026 in New York, SAP announced a storefront Model Context Protocol (MCP) server for SAP Commerce Cloud, together with a set of agentic shopping capabilities. The point is not that SAP is doing AI in commerce; every vendor is. The point is how: SAP chose an open protocol that any AI agent can use, instead of limiting agentic shopping to its own assistant.

We build and run MCP servers ourselves, for example for SAP Sales and Service Cloud V2, so we read the announcement with some practical experience of what such a server needs underneath. Below is what SAP has confirmed, what is still open, and what we recommend doing now.

TL;DR: SAP’s storefront MCP server lets AI agents work directly with a Commerce Cloud storefront: find products, check prices and stock, manage carts and complete purchases. It is not a chatbot but an open interface for any MCP-compatible agent, including third-party assistants. SAP announced it in January 2026 for Q2 and listed it in its Q2 2026 release. The B2B use cases (reordering, procurement agents) pay off sooner than B2C. The preparation that matters most is product data quality.

What did SAP announce at NRF 2026?

SAP’s announcements in January 2026 had three parts that belong together.

First, the storefront MCP server for SAP Commerce Cloud. SAP describes it as a way to make storefronts “intelligible to AI”, connecting products, pricing, inventory and promotions to AI-driven discovery and shopping, on the retailer’s own site and on platforms such as ChatGPT (SAP News, January 2026). SAP planned it for Q2 availability and positions it as a first step towards supporting several agentic protocols: MCP, OpenAI’s Agentic Commerce Protocol (ACP) and Google’s Universal Commerce Protocol (UCP) (SAP News: Agentic AI is reshaping commerce).

Second, a Shopping Agent. SAP presented it as something that can be combined with the Digital Service Agent into one conversational assistant covering discovery, purchase and after-sales support (SAP News: For retailers, agentic commerce is here). We describe the service side in our post on the Digital Service Agent in SAP Service Cloud V2.

Third, the delivery. In its Q2 2026 release summary, SAP lists “conversational AI shopping through a model context protocol (MCP) server”: AI assistants can query real-time product information, provide inventory updates, manage carts and complete transactions within chat or voice interfaces (SAP News, July 2026).

What is MCP (Model Context Protocol)?

The Model Context Protocol is an open standard, introduced by Anthropic in 2024, that defines how AI models connect to external data and tools (modelcontextprotocol.io). Compare it to REST: a common interface that lets systems talk to each other without a custom integration for every pair.

Without such a standard, every AI assistant that wants to work with a commerce platform needs its own connector, and each connector works only with one vendor’s AI. An MCP server instead exposes a set of capabilities that any compliant agent can discover and use. One MCP server for Commerce Cloud means that ChatGPT, Claude, Gemini, Joule and future MCP-compatible agents can work with your storefront through the same interface. You build the integration once.

The protocol defines three types of capabilities:

• Tools: actions the agent can perform (search products, add to cart, apply a coupon)
• Resources: data the agent can read (product catalogue, order history)
• Prompts: predefined interaction templates (product comparison, reorder)

For SAP this is a notable decision: it bets on an open standard rather than tying agentic commerce to Joule alone.

How does the Commerce Cloud MCP server work?

SAP has published the capabilities, not a full technical specification, so the following separates what is confirmed from what follows from how MCP servers are usually built.

Confirmed by SAP: agents can query real-time product information and inventory, manage carts and complete transactions, and they can come from the retailer’s own experience or from third-party assistants.

Architecture, as we expect it: the MCP server sits between external AI agents and Commerce Cloud’s existing commerce services (the same ones the OCC REST APIs expose). It does not replace your storefront and runs next to your Composable Storefront or custom frontend:

User → AI agent (ChatGPT, Claude, Joule, ...) → MCP → Commerce Cloud MCP server → Commerce services → Commerce Cloud backend

What to verify in your project: how authentication works when an agent acts for a customer (it must not see more than the customer would, for example no wholesale prices for a B2C login), which B2B functions are covered (requisition lists, approval workflows, contract prices), and how agent actions are logged. These are the questions we would put to SAP before any pilot, and they are the same ones we answer when we build MCP servers for other SAP products.

Agentic AI shopping: what could it look like?

Two illustrative scenarios show what the confirmed capabilities make possible. The single steps (search, filter, cart, checkout) already exist in Commerce Cloud; MCP lets an agent chain them.

B2C scenario: the AI personal shopper

A customer tells their AI assistant: “I’m training for a marathon in May. I need road running shoes, a hydration vest in a women’s fit and energy gels without caffeine. Budget around CHF 400.”

The agent, connected to your storefront via MCP:

1. Searches for road running shoes and filters by availability and the customer’s usual size
2. Picks three options within the budget and compares cushioning, weight and ratings
3. Searches hydration vests with a women’s fit and keeps the total within budget
4. Finds caffeine-free energy gels using the product attributes
5. Assembles a cart and shows the customer a summary with the total price
6. Waits for the customer’s approval before checkout

The customer swaps one item and approves. What used to be half an hour of browsing across several categories takes a few minutes.

B2B scenario: the procurement agent

A procurement manager at a manufacturer reorders consumables (cutting fluids, abrasives, safety equipment) every two weeks. Today that means logging into the portal, checking stock, comparing with the last order, adjusting quantities and routing the order for approval.

With an MCP-connected agent:

1. The agent watches stock levels in the company’s ERP, for example SAP S/4HANA
2. When a threshold is reached, it checks current prices and availability in the supplier’s B2B storefront via MCP
3. It compares contract prices with the last three orders and flags deviations
4. It prepares a draft order with the negotiated contract terms
5. It sends the budget owner a summary for approval: “Reorder matches the usual pattern. Two items changed price. Total is within budget.”

The budget owner approves, the agent completes the order through the normal checkout, and every step is traceable. Whether your supplier’s storefront supports this depends on the B2B scope of the MCP server, which is exactly what to check first. How the ERP side of such a flow fits together is covered in SAP Commerce Cloud and S/4HANA Public Cloud for B2B.

What does this mean for B2B commerce?

B2B is where agentic commerce pays off fastest. Buying processes repeat, the data is structured, and approval workflows already exist in Commerce Cloud.

Automated reordering is the obvious starting point. Most B2B customers buy largely the same products every cycle. An agent can learn the pattern, watch stock signals and prepare orders. The human stays in the loop for approval, but reviewing a summary replaces clicking through a portal.

Approvals with context get better. Instead of a bare purchase order, the approver gets price trends, alternatives that meet the same specification and delivery dates. Decisions become faster and better founded.

Comparing suppliers is the longer-term prospect. If several suppliers expose MCP servers, a procurement agent could compare prices, availability and lead times across them. That depends on broad adoption, but the protocol makes it possible in principle.

Contract compliance is a natural fit as well: the agent checks each order against negotiated terms and flags deviations before submission. Many procurement teams do this manually today, or not at all.

What does this mean for B2C commerce?

In B2C the change is mostly in the experience, and the return is harder to measure early on.

Conversational shopping replaces browse, filter and sort. Customers describe what they need and the agent translates intent into products. This helps most with complex purchases (electronics, DIY, specialty food) where customers do not know the right category or specification.

Recommendations go beyond “customers who bought X also bought Y”. An agent with access to purchase history, stated preferences, the current cart and the catalogue can reason about compatibility and budget.

After-sales support changes when the agent knows the order. “My order hasn’t arrived” leads to a status and tracking check without the customer searching for the order page. This is where the Shopping Agent and the Digital Service Agent are meant to meet.

Promotions can be matched to the situation: a customer buying a coffee machine gets an offer for compatible capsules, delivered within the conversation.

The open question in B2C is the effect on conversion. There are no reliable figures for MCP-based shopping yet, so we would treat any pilot as a measured experiment with a clear baseline.

How should you prepare your Commerce Cloud for AI agents?

Whether you switch the MCP server on soon or wait, the preparation is the same. None of it is wasted, because it also improves your normal online shop.

Product data quality

This is the most important factor. An agent is only as useful as the data it can read. If a product is described as “Blue Widget, SKU 12345” without attributes, specifications or a meaningful category, the agent has nothing to work with.

What good looks like:

• Natural-language product descriptions: what the product is, who it is for and how it differs from alternatives. Agents read language, not SEO keyword lists.
• Complete attributes: dimensions, materials, compatibility, certifications, maintained in Commerce Cloud’s classification system. Empty attributes are invisible attributes.
• A consistent category taxonomy: a clean category tree helps agents in the same way it helps people. Legacy categories, duplicate nodes and misfiled products confuse both.
• Accurate, current inventory: an agent that puts out-of-stock items in the cart and then fails at checkout destroys trust. For agentic commerce, inventory accuracy is a prerequisite.

API readiness

The MCP server builds on Commerce Cloud’s commerce services. If you already run a headless setup or Composable Storefront, your OCC layer is probably in good shape. If you are on an older Accelerator storefront that exposes little through OCC, there is work to do. Check product search with facets, cart operations, customer authentication, order history and tracking, and promotions.

If you still run SAP Commerce on-premise, note that SAP positions the MCP server as a Commerce Cloud capability, and mainstream maintenance for the on-premise release ended on 31 July 2026. We describe the options in SAP Commerce on-premise support has ended: now what?.

Content and media

Agents are increasingly multimodal. Make sure product images have meaningful alt text (not “product-image-1.jpg”), several angles and context shots, and a size reference where it matters. Your product media is how the agent shows results to the customer.

How does the competition compare?

Every major commerce platform is moving towards agentic commerce, and open protocols are becoming the common ground.

Shopify co-developed the Universal Commerce Protocol with Google, announced at NRF in January 2026 (Google Developers Blog), and provides its own MCP servers for storefront, cart and checkout.

Salesforce builds its commerce agents on Agentforce, its own agent platform, which fits best if the rest of your landscape is Salesforce.

Adobe Commerce offers AI features for search, recommendations and catalogue management within the Adobe ecosystem.

What sets SAP apart is less the protocol itself, since others use MCP too, than the combination: an open interface for external agents on top of a commerce platform that is closely tied to the ERP. For B2B companies on SAP S/4HANA, that means an agent can work with contract prices, availability and order data that come from the same system of record. How well this works in practice depends on the scope SAP delivers and on how cleanly Commerce and ERP are integrated in your landscape.

Readiness checklist

SAP’s roadmap for the MCP server beyond the Q2 2026 delivery is not public in detail; check the SAP Roadmap Explorer and your SAP contact for the current state. Regardless of timing, these steps make sense now:

• Audit product data quality: run a completeness report on descriptions, attributes and classifications, starting with your best-selling products.
• Review OCC API coverage: make sure all storefront functions are available through the API, not just the UI. Test search, cart, checkout and order flows programmatically.
• Clean up the category taxonomy: remove duplicate categories and fix misfiled products. What confuses a human confuses an agent.
• Verify inventory accuracy: compare Commerce Cloud stock with warehouse data and close synchronisation gaps.
• Rewrite key product descriptions: descriptive and complete, with use cases, compatibility and comparison points.
• Check your platform: if you still run SAP Commerce on-premise, plan the move to SAP Commerce Cloud.
• Brief your team: e-commerce, IT and business should share a sober picture of what agentic commerce means for your business.
• Pick one pilot use case: B2B reordering or B2C conversational search. The best candidates are frequent, repetitive processes with structured data.

---

We have delivered SAP Commerce Cloud projects for years, and we build MCP servers for SAP products ourselves. For a broader view of SAP’s agentic AI beyond commerce, read Agentic AI at SAP: what it means for sales; for how we compare Swiss Commerce partners, see the best SAP Commerce Cloud partners in Switzerland. If you are planning your Commerce Cloud AI roadmap, let’s talk.

Frequently asked questions

What is SAP’s Commerce Cloud MCP server?

It is a server based on the open Model Context Protocol that makes an SAP Commerce Cloud storefront usable for AI agents. According to SAP, assistants can query real-time product information and inventory, manage carts and complete transactions from chat or voice interfaces. The agent can run in the retailer’s own experience or in a third-party assistant such as ChatGPT or Perplexity.

When is the MCP server available?

SAP announced the storefront MCP server at NRF in January 2026 with availability planned for Q2 2026, and listed it in its SAP Customer Experience Q2 2026 release summary in July 2026. Confirm the scope and commercial terms for your tenant with SAP before you plan a project around it.

What is agentic AI in e-commerce?

AI agents that carry out multi-step shopping tasks towards a goal: research products, compare options, check availability, apply promotions, assemble a cart and route it to checkout. Unlike a chatbot that answers single questions, the agent chains the steps itself, while a human approves the purchase and the payment.

Do I need to rebuild my storefront for this?

No. The MCP server is an additional interface for AI agents on top of Commerce Cloud’s commerce services, next to your existing Composable Storefront or custom frontend. What usually needs work is not the storefront but the data behind it: product attributes, taxonomy, inventory accuracy and complete OCC API coverage.