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

This one caught us off guard. Not because SAP is doing AI in commerce — everyone is — but because of how they’re doing it. At NRF 2026 in New York, SAP quietly announced a Model Context Protocol server for Commerce Cloud storefronts alongside a set of agentic AI shopping capabilities built on top of it.

No third-party guides exist yet. No analyst write-ups. No implementation blogs. We’ve been pulling apart the announcement, cross-referencing with SAP’s technical documentation, and running early experiments in our own Commerce Cloud sandbox. Here’s what we’ve found.

TL;DR: SAP is building an MCP server that lets AI agents interact directly with Commerce Cloud storefronts — browsing products, managing carts, checking inventory, and completing purchases. This is not a chatbot. It’s an open protocol that turns your storefront into an API surface for any MCP-compatible AI agent. The B2B implications (automated procurement, reorder agents) are more immediate than B2C. Available Q2 2026. Start preparing your product data now.

What SAP Announced at NRF 2026

NRF 2026 ran January 12-14 in New York. SAP’s announcement had three components that, taken together, represent a meaningful shift in how e-commerce platforms will interact with AI systems.

First, the Storefront MCP Server. This is an implementation of the Model Context Protocol — the open standard created by Anthropic — that sits on top of Commerce Cloud’s OCC (Omnichannel Commerce) APIs. It exposes product catalogs, category trees, pricing, inventory, cart operations, and order history to any MCP-compatible AI agent. Think of it as a structured translator between your storefront and the growing ecosystem of AI assistants.

Second, agentic shopping capabilities. SAP demonstrated AI agents that could perform multi-step shopping tasks autonomously: researching products across categories, comparing specifications, checking real-time availability, applying promotions, and assembling a complete cart — all without human intervention until the final purchase approval. These aren’t canned demo flows. They’re built on the MCP server’s real-time connection to live Commerce Cloud data.

Third, AI-powered product discovery. Moving beyond keyword search, SAP showed natural language queries like “I need running shoes for wet conditions under CHF 200” that return semantically relevant results, not just keyword matches. This uses Commerce Cloud’s product data enriched with vector embeddings that understand product relationships and attributes.

This builds on SAP’s broader AI push. SAP has deployed 350 AI features with 2,400+ Joule skills across its cloud portfolio (SAP News Center, 2026). Commerce Cloud is getting the next wave.

What Is MCP (Model Context Protocol)?

If you’re not deep in the AI tooling world, MCP deserves a quick explanation. It matters more than most people realise.

The Model Context Protocol is an open standard — originally developed by Anthropic — that defines how AI models connect to external data sources and tools. Think of it the way you think about REST APIs: a common interface that lets different systems talk to each other without custom integrations for each pair.

Before MCP, every AI assistant that wanted to interact with a commerce platform needed custom-built connectors. Shopify had its own AI integration. Adobe had its own. Salesforce had its own. Each required dedicated engineering effort, and each worked only with that vendor’s AI tools.

MCP changes this. An MCP server exposes a set of capabilities — “tools” in MCP terminology — that any compliant AI agent can discover and use. A single MCP server for Commerce Cloud means that Claude, GPT-based agents, Gemini, Joule, and any future MCP-compatible AI can interact 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 (add to cart, search products, apply coupon)
• Resources: Data the agent can read (product catalog, order history, customer profile)
• Prompts: Pre-built interaction templates (product comparison, reorder workflow)

SAP’s announcement puts Commerce Cloud among the first enterprise commerce platforms to implement MCP at the storefront level. That’s significant, because it means SAP is betting on an open standard rather than locking AI capabilities into Joule alone.

How the Commerce Cloud MCP Server Works

Based on what SAP has shared and our own early testing, here’s the architecture.

The MCP server sits as a layer between Commerce Cloud’s existing OCC REST APIs and external AI agents. It doesn’t replace your storefront. It doesn’t require changes to your existing Spartacus or Composable Storefront implementation. It runs alongside everything you already have.

The server exposes Commerce Cloud capabilities as MCP tools:

Product discovery tools — search by natural language query, browse by category, get product details, check real-time inventory and pricing. The natural language search uses vector embeddings generated from your product catalog’s attributes, descriptions, and classification data.

Cart management tools — create cart, add items, remove items, update quantities, apply promotion codes. The agent handles session management and can maintain multiple carts (useful for B2B scenarios where a procurement agent might be assembling orders for different cost centres).

Order tools — retrieve order history, check order status, initiate returns. In B2B contexts, this extends to requisition lists, approval workflows, and contract-based pricing lookups.

Customer context tools — access saved addresses, payment methods, wish lists. All scoped to the authenticated user’s permissions and data access rights.

Authentication flows through Commerce Cloud’s existing OAuth layer. An AI agent authenticating via MCP receives the same permissions and data access as the user it represents. No backdoor access, no elevated privileges. If a customer can’t see wholesale pricing in your storefront, an AI agent acting on their behalf can’t see it through MCP either.

The data flow looks like this:

User → AI Agent (Claude, Joule, etc.) → MCP Protocol → Commerce Cloud MCP Server → OCC APIs → Commerce Cloud Backend

Every request traces back to an authenticated session. Every action is logged in Commerce Cloud’s standard audit trail. This isn’t a separate system — it’s a new interface to the existing one.

Agentic AI Shopping: What It Looks Like

Let’s make this concrete with two scenarios that SAP demonstrated and that we’ve been exploring in our sandbox environment.

B2C Scenario: The AI Personal Shopper

A customer tells their AI assistant: “I’m training for a marathon in May. I need new running shoes for road running, a hydration vest that fits women’s sizes, and some energy gels — nothing with caffeine. Budget around CHF 400 total.”

The AI agent, connected to your Commerce Cloud storefront via MCP, does the following without further prompting:

1. Searches the product catalog for road running shoes, filters by availability and customer’s previously purchased size
2. Identifies three options within the budget allocation, comparing cushioning, weight, and customer ratings
3. Searches hydration vests with women’s fit filter, cross-references with the running shoe selection to stay within total budget
4. Finds caffeine-free energy gels, checks compatibility notes in product attributes
5. Assembles a cart with recommended products, shows the customer a summary with total price
6. Waits for human approval before completing checkout

The customer reviews, swaps one item, approves. Done. What would have been 30 minutes of browsing, filtering, comparing, and checking reviews across multiple categories happened in under two minutes.

This isn’t science fiction. The individual steps — search, filter, cart management, checkout — all exist in Commerce Cloud today. MCP provides the protocol for an AI agent to chain them together autonomously.

B2B Scenario: The Procurement Agent

A procurement manager at a manufacturing company has a recurring need: reorder consumable supplies (cutting fluids, abrasives, safety equipment) every two weeks. Currently, this takes an hour of logging into the portal, checking stock, comparing to the last order, adjusting quantities, and routing through approval.

With an MCP-connected AI agent:

1. The agent monitors inventory levels through integration with the company’s ERP (SAP S/4HANA via BTP)
2. When thresholds trigger, it accesses the B2B storefront via MCP to check current pricing and availability
3. Cross-references the contract pricing in Commerce Cloud with the last three orders to flag anomalies
4. Assembles a draft purchase order, applies the negotiated contract terms automatically
5. Routes to the budget owner for approval with a summary: “Reorder matches previous pattern. Two items have price changes. Total is 3% under budget.”

The budget owner reviews a clean summary, approves with one click. The agent completes the order through the standard checkout flow, and the audit trail captures every step.

This B2B scenario is, frankly, where the real money is. We already have clients spending 40+ hours per month on routine reordering. Automating that through an intelligent agent that understands both the internal needs and the supplier’s catalog is a tangible, measurable efficiency gain.

What This Means for B2B Commerce

B2B is where agentic commerce delivers the fastest return. The buying processes are repetitive, the data is structured, and the approval workflows already exist in Commerce Cloud.

Automated reordering is the lowest-hanging fruit. Most B2B customers buy the same 80% of products every cycle. An AI agent can learn the pattern, monitor inventory signals, and assemble orders proactively. The human stays in the loop for approval, but the 45 minutes of clicking through a portal drops to 2 minutes of reviewing a summary.

Approval workflows with AI context become smarter. Instead of a budget owner receiving a bland purchase order for approval, the AI agent attaches context: price trend analysis, alternative products that meet the same spec at lower cost, delivery timeline comparison. The approver makes a better decision faster.

Supplier comparison across storefronts is the longer-term play. If multiple suppliers expose MCP servers, a procurement agent could compare pricing, availability, and lead times across suppliers in real time. This is still theoretical — it requires multiple suppliers to adopt MCP — but the protocol makes it architecturally possible in a way it wasn’t before.

Contract compliance monitoring is another natural fit. The agent can validate that every order adheres to negotiated terms, flag deviations before they’re submitted, and maintain a running log of contract utilisation. Procurement teams currently do this manually or not at all.

The global B2B e-commerce market continues to dwarf B2C in transaction volume. Gartner estimates that by 2028, 60% of B2B sales transactions will occur through digital channels (Gartner, 2025). Agentic commerce doesn’t just digitise these transactions — it automates the decision-making around them.

What This Means for B2C Commerce

B2C is where the experience innovation happens, even if the ROI is harder to pin down immediately.

Conversational shopping replaces the browse-filter-sort paradigm. Instead of navigating a category tree, customers describe what they need in natural language. The AI agent handles the translation from intent to product results. This is especially powerful for complex purchase decisions — electronics, home improvement, specialty food — where customers don’t know the exact product category or technical specification they need.

AI-curated recommendations move beyond “customers who bought X also bought Y” collaborative filtering. An MCP-connected agent has access to the customer’s full purchase history, stated preferences, current cart, and the complete product catalog. It can reason about complementary products, compatibility, and budget constraints in ways that a traditional recommendation engine cannot.

Post-purchase support transforms when the agent has full order context. “My order hasn’t arrived” triggers an agent that checks the order status, tracking information, and delivery estimates without the customer navigating to an order history page or calling support. Returns become “I need to return the blue jacket from my last order” — the agent knows which order, which item, and can initiate the return flow immediately.

Personalised promotions get more targeted. Rather than showing the same banner to everyone, the AI agent can evaluate a customer’s browsing context, purchase history, and current cart in real time and surface relevant offers. “You’re buying a coffee machine — we have a 20% discount on compatible capsules this week” delivered through the agent’s natural conversation flow.

The key question for B2C is conversion rate impact. Early data from conversational commerce implementations — not MCP-specific, but similar patterns — suggests a 15-25% increase in conversion for customers who engage with AI shopping assistants compared to traditional browse-and-buy flows. But this data is still thin and mostly from early adopters.

How to Prepare Your Commerce Cloud for AI

Regardless of whether you plan to enable the MCP server on day one or wait for it to mature, the preparation work is the same. And none of it is wasted even if you never adopt agentic commerce, because it all improves your standard e-commerce operations.

Product Data Quality

This is the single most important factor. An AI agent is only as helpful as the data it can access. If your product descriptions say “Blue Widget - SKU 12345” with no attributes, no specifications, and no meaningful categorisation, the agent has nothing to work with.

What “good” looks like for MCP readiness:

• Rich, natural language product descriptions — not keyword-stuffed marketing copy, but genuine descriptions of what the product is, who it’s for, and how it compares to alternatives. AI agents parse natural language, not SEO patterns.
• Complete attribute sets — dimensions, materials, compatibility, certifications. Structured data in Commerce Cloud’s classification system is what the MCP server exposes to agents. Empty attributes are invisible attributes.
• Consistent category taxonomy — a clean, logical category tree helps agents navigate your catalog the same way it helps human customers. If your taxonomy is a mess of legacy categories, duplicate nodes, and miscategorised products, the agent will struggle.
• Accurate, real-time inventory data — an agent that adds out-of-stock items to a cart and then fails at checkout destroys trust. Inventory accuracy isn’t a nice-to-have for agentic commerce. It’s a prerequisite.

API Readiness

The MCP server builds on Commerce Cloud’s OCC APIs. If you’ve been using headless or Composable Storefront, your APIs are probably in good shape. If you’re on an older Accelerator storefront that doesn’t expose much through OCC, there’s work to do.

Check that your OCC layer covers: product search with faceted filtering, cart CRUD operations, customer authentication flows, order history and tracking, promotion and coupon application.

If you’re still running SAP Commerce on-prem, the MCP server is another reason to plan your cloud migration. The server requires Commerce Cloud’s managed infrastructure — it’s not available for on-prem installations.

Content and Media

AI agents today are primarily text-based, but multimodal agents that can process images are coming fast. Ensure your product images have:

• Meaningful alt text (not “product-image-1.jpg”)
• Multiple angles and context shots
• Size/scale references where relevant

Your product media becomes the agent’s way of showing results to the customer. Poor media means poor agent-assisted experiences.

The Competitive Landscape

SAP isn’t operating in a vacuum. Every major commerce platform is building AI capabilities, but the approaches differ significantly.

Shopify Sidekick focuses on the merchant side — helping store owners manage their business with AI rather than helping shoppers buy. It handles inventory analysis, marketing copy generation, and business insights. For the shopper-facing side, Shopify’s Shop app has conversational features, but there’s no MCP-level open protocol for third-party AI agents to interact with Shopify storefronts directly.

Adobe Sensei in Commerce provides AI-powered product recommendations, visual search, and intelligent catalog management. Adobe has strong AI capabilities but has not announced MCP support. Their AI tools are mostly internal to the Adobe ecosystem — AI features that work within Adobe Commerce, not an open protocol for external agents.

Salesforce Einstein for Commerce offers predictive product sorting, AI-powered search, and personalised recommendations. Salesforce has Agentforce — their agentic AI platform — which includes commerce capabilities. The approach is tightly coupled to the Salesforce ecosystem, using their own agent framework rather than an open protocol.

What makes SAP’s approach different is the bet on MCP as an open standard. By implementing MCP rather than a proprietary agent framework, SAP is saying: we want any AI agent to be able to interact with Commerce Cloud, not just Joule. That’s a meaningful architectural decision. It means customers aren’t locked into SAP’s AI tools to get agentic commerce capabilities.

Whether this openness becomes a competitive advantage depends on MCP adoption across the industry. If MCP becomes the standard protocol for AI-to-application communication — which the trajectory suggests — SAP’s early adoption is a strong position. If the market fragments into proprietary agent frameworks, it’s a bet that may not pay off.

Our take: MCP is winning. The protocol has momentum, major AI labs support it, and the enterprise software ecosystem is converging on it. SAP is on the right side of this one.

Timeline and Readiness Checklist

Based on SAP’s announcements and our conversations with the Commerce Cloud product team, here’s what we expect:

Q2 2026 (April-June): MCP server available in early adopter programme. Limited to product discovery and cart management tools. Requires Commerce Cloud 2211 or later.

Q3 2026: Expanded tool set including order management, customer profile access, and B2B-specific features (requisition lists, approval routing). General availability likely.

Q4 2026 and beyond: Advanced capabilities — multi-agent orchestration, cross-system integration (Commerce + S/4HANA via BTP), predictive reordering.

Your Readiness Checklist

Here’s what to tackle now, regardless of your MCP timeline:

• Audit product data quality — run a completeness report on product descriptions, attributes, and classifications. Target >90% attribute fill rate for your top 500 products.
• Review your OCC API coverage — ensure all storefront functions are accessible via API, not just through the UI layer. Test search, cart, checkout, and order flows programmatically.
• Clean up your category taxonomy — eliminate duplicate categories, fix miscategorised products, ensure logical navigation paths. What confuses a human will confuse an agent.
• Verify inventory accuracy — compare Commerce Cloud inventory records against actual warehouse data. Fix synchronisation gaps. Real-time accuracy matters more for agents than for human shoppers.
• Update product descriptions for natural language — rewrite top-selling product descriptions to be descriptive and complete, not just marketing copy. Include use cases, compatibility, and comparison points.
• Assess your Commerce Cloud version — MCP server requires 2211+. If you’re on an older version, plan the upgrade. If you’re on-prem, this is another trigger for cloud migration.
• Brief your team — ensure your e-commerce, IT, and business stakeholders understand what agentic commerce means and what’s coming. Avoid the hype, focus on the practical use cases relevant to your business.
• Identify one pilot use case — pick either B2B reordering automation or B2C conversational search as your first MCP experiment. The best candidates are high-volume, repetitive processes with structured data.

We’ve been building on SAP Commerce Cloud for years, and we’ve been working with MCP servers across the SAP ecosystem. The intersection of these two is exactly where we think the next wave of commerce innovation is heading. If you’re planning your Commerce Cloud AI strategy — or just trying to figure out what NRF 2026 means for your roadmap — let’s talk.

Frequently Asked Questions

What is SAP’s Commerce Cloud MCP server?

It’s an implementation of the Model Context Protocol that exposes Commerce Cloud storefront data — products, categories, cart, orders — to AI agents. Any MCP-compatible AI assistant (Claude, Joule, GPT-based agents, Gemini) can connect to it and perform shopping tasks on behalf of users. It builds on Commerce Cloud’s existing OCC APIs and runs alongside your current storefront without requiring changes to your customer-facing experience.

When will the MCP server be available?

SAP announced availability starting Q2 2026 at NRF in January 2026. The early adopter programme covers product discovery and cart management. Full general availability with B2B features and order management is expected Q3 2026. The exact timeline depends on your Commerce Cloud version (2211+ required) and region. Check with your SAP account executive for your specific availability window.

What is agentic AI in e-commerce?

AI agents that can autonomously perform multi-step shopping tasks. Not a chatbot that answers product questions — an agent that can research products, compare options across categories, check real-time inventory, apply promotions, assemble a cart, and route through checkout. Human approval remains at key decision points (purchase confirmation, payment authorisation). The distinction from traditional AI is autonomy: the agent pursues a goal through multiple steps rather than responding to individual commands.

Do I need to rebuild my storefront for this?

No. The MCP server runs as a separate layer on top of Commerce Cloud’s OCC APIs. Your existing Spartacus, Composable Storefront, or custom frontend continues to operate exactly as before. The MCP server adds a new interface for AI agents to interact with the same backend — it doesn’t replace or modify your current customer-facing experience. The main prerequisite is that your Commerce Cloud instance is on version 2211 or later and your OCC APIs are properly configured.