Key Takeaways
- On June 29, 2026, Microsoft launched a Dynamics 365 Commerce MCP server in public preview, letting AI agents directly call everything from product discovery through checkout and order lookup.
- MCP (Model Context Protocol) is an open standard introduced by Anthropic in 2024; Microsoft used it to open up its headless commerce engine to an agent channel without replatforming.
- For retailers, the biggest implication is being able to serve the exact same pricing, inventory, and promotions as their website through agents on ChatGPT, Copilot, and Claude.
Microsoft opens Dynamics 365 Commerce to AI agents
Discover how the Dynamics 365 Commerce MCP server enables AI agents to power product discovery, checkout, inventory, and retail operations.
www.microsoft.comWhether an AI agent can pull its weight in retail comes down to how many back-end systems it can actually reach. Until now, connecting an agent to commerce meant building custom integrations for every data source and channel, an approach that is brittle, slow, and hard to scale.
On June 29, 2026, Microsoft answered that problem by announcing the public preview of the Dynamics 365 Commerce MCP server. It was unveiled at NRF 2026, retail's largest trade show, and the company frames it as the foundation for agentic commerce. Retail-specific business logic, including product discovery, inventory availability, pricing and discounts, cart, checkout, and order lookup, is now exposed as standardized tools any compatible AI agent can call directly.
MCP (Model Context Protocol) here refers to an open standard for connecting AI agents to data systems and business logic. A platform exposes its capabilities once, and supported agents can then call them across multiple surfaces in real time. Microsoft used this mechanism to open the very same headless commerce engine that powers its storefronts to agents, with no need to rebuild the stack.
Tools designed around intent, not raw APIs
The technically important part of this announcement is that the Dynamics 365 Commerce MCP server is built around customer intent rather than simply re-exposing individual APIs. An agent can discover products, check availability, apply discounts, build a cart, and select fulfillment options inside a single interaction.
The public preview ships with roughly 15 tools that cover the core of the shopping journey.
| Category | Key tools | What it does |
|---|---|---|
| Discovery & inventory | search_products / get_product_by_id / search_store_inventory | Keyword and attribute search, full product detail, on-hand inventory including nearby stores |
| Cart & checkout | create_cart / add_product_to_cart / update_cart_address / get_cart_delivery_options / create_payment_link | Build carts, add items, set address, fetch delivery options, generate a Pay by Link (Adyen) checkout URL |
| Promotions | apply_coupon_code / remove_coupon_code | Apply and remove coupons (HQ-configured promotions auto-apply) |
| Order & account | get_order_details / get_order_history / get_saved_delivery_addresses | Anonymous and authenticated order lookup, cross-channel order history, saved addresses |
What stands out is that these tools return the exact same data shape that the website and store POS receive. Promotions and tax are not a separate engine either; the same promotion logic configured in headquarters auto-applies in the agent channel. For payments, rather than letting the agent handle card data, the design uses Pay by Link (Adyen) to generate a secure checkout URL the shopper completes. That is a deliberate decision to keep payment details out of the agent, and a pragmatic compliance choice.
The server itself runs as a Microsoft-managed, first-party endpoint on the Commerce Scale Unit (CSU) that powers Dynamics 365 Commerce storefronts and POS. It requires CSU version 10.0.48 or later, and lets retailers focus on differentiated commerce experiences without building or maintaining their own MCP infrastructure.
It runs on ChatGPT, Copilot, and Claude as-is
What makes this practical for retailers is that it is agnostic about which agent connects. According to Microsoft's documentation, the Dynamics 365 Commerce MCP server exposes the standard MCP protocol over HTTP and accepts connections from compliant clients including Microsoft Copilot Studio, OpenAI's ChatGPT, Anthropic's Claude Desktop, Azure AI Foundry, or your own agent host. The "build once, extend across channels" philosophy takes concrete shape here.
On the authentication side, agents always call tools under a user context using a Microsoft Entra ID bearer token. Anonymous guests can browse the public catalog, pricing, and inventory and use guest carts and Pay by Link checkout, while signed-in C2 (consumer) shoppers unlock customer-specific pricing, saved addresses, order history, and order tracking. Once a shopper signs in, that same identity flows all the way through to the commerce engine, so they see one order history spanning store, web, call center, and agent. That is what Microsoft means by omnichannel in practice.
Early adoption is already underway. Michael Hill, one of Australia's leading jewelry retailers, is running early tests, and its group retail and omnichannel systems manager put it this way.
It creates a practical way to introduce AI-driven innovation at the pace retail demands, simply by layering intelligence over existing commerce platforms.
Microsoft partner AMICIS Solutions built a voice-driven commerce agent on the server, demonstrating store associates searching the catalog, managing carts, looking up customers, and handling everyday POS tasks through natural language.
Reach extends to store operations and autonomous agents
This release goes well beyond consumer-facing conversational commerce. Microsoft is also pushing in two further directions: operational support for store associates and autonomous agents that act without a human in the loop.
In stores, associates can complete returns, exchanges, and order lookups by voice instead of memorizing multi-screen POS flows, and clienteling agents can pull together purchase history, preferences, and loyalty data for personalized service. On the autonomous side, agents are envisioned to sense restocking needs from B2B demand signals and recommend or trigger reorders, detect price drops and reach out to customers with personalized offers, or reposition inventory across locations based on predicted demand.
Microsoft also outlines pairing this Commerce MCP server with the Dynamics 365 ERP MCP server it shipped earlier in November 2025. Commerce handles the selling side (discovery, pricing, checkout, order management) while ERP handles what comes before and after (merchandising, demand planning, procurement, inventory allocation, fulfillment, financials), giving agents access to the whole value chain from planning to selling to settlement.
MCP is Anthropic's standard. Where does Microsoft sit?
It is essential to recognize that MCP is not Microsoft's own invention. MCP is an open standard Anthropic introduced in November 2024 to standardize how AI assistants connect to external tools and data. Within roughly a year, major providers including OpenAI and Google DeepMind adopted it, making it the de facto standard for connecting agents to tools. In December 2025, Anthropic donated MCP to the Linux Foundation's Agentic AI Foundation, moving it to neutral governance.
Standardization in agentic commerce is happening along several parallel tracks. Shopify and Google co-developed the Universal Commerce Protocol (UCP) that defines the transaction flow, and Microsoft is among its backers. OpenAI and Stripe lead the Agentic Commerce Protocol (ACP) for checkout and payment, and Salesforce announced MCP support in Agentforce 3 back in June 2025. A division of labor is emerging: MCP as the discovery-and-dialogue standard, UCP and ACP as transaction-flow standards.
Within that, Microsoft's move is to wire its commerce engine directly into MCP, the discovery standard, and layer retail capabilities on top of existing agents including ChatGPT and Claude. Comparing the standards clarifies how each is positioned.
| Standard | Led by | Primary role |
|---|---|---|
| MCP | Anthropic (now Agentic AI Foundation) | Connecting agents to tools and data; the standard for discovery and dialogue |
| UCP | Shopify / Google | Defining the transaction flow: cart, checkout, payment, and post-purchase |
| ACP | OpenAI / Stripe | Checkout sessions and secure transmission of payment data |
What retailers should prepare for
The implications for ecommerce and retail operators are clear. First, rather than "yet another channel," this means a new option to extend selling into the agent surface without rebuilding your existing commerce foundation. Avoiding the drift where pricing, inventory, and promotions diverge from your other channels is not trivial from an operations and support-ticket standpoint.
Second, structuring your data so agents can find the right products correctly becomes a new discipline alongside search engine optimization. The MCP server returns structured product, inventory, and pricing data, and how accurately the tools interpret intent depends on the quality of that source data. Microsoft's emphasis on continuous evaluation reflects that whether an agent selects the right tool and passes the right parameters in production is the heart of reliability.
Third, licensing deserves attention. While there is no extra license fee for the MCP server during preview, Microsoft has reportedly signaled it will become part of a higher Dynamics 365 Commerce premium tier once it reaches general availability. It is realistic to plan deployments on that assumption.
Conclusion
The Dynamics 365 Commerce MCP server is a clear move by Microsoft to connect its commerce engine to the agent economy on top of Anthropic's neutral MCP standard. The same engine that powers your website runs inside ChatGPT, Copilot, and Claude as-is, and that continuity is the essence of this announcement.
What to watch next is the timing of the move from public preview to general availability, and how far the pairing with the ERP MCP server can carry "planning to settlement" through in real operations. How interoperability with transaction standards like UCP and ACP gets sorted out will also shape how much each operator should invest in which standard. Rethinking product data and channel design for agent-led purchasing is no longer a topic anyone can defer.
