Microsoft has opened the public preview of its Dynamics 365 Commerce MCP server, giving retailers a managed endpoint that lets AI agents directly interact with checkout systems for the first time. Available now on Commerce Scale Unit 10.0.48, the June 29, 2026 release marks a significant step toward fully autonomous shopping experiences, where AI assistants can handle product lookups, tax calculations, payment processing, and order placement without human intervention.
This isn’t just another API. The MCP server – short for Microsoft Commerce Platform – acts as a dedicated, Microsoft-managed gateway that translates intentions from approved AI models into secure, transactional actions inside the Dynamics 365 Commerce engine. For retailers already invested in Microsoft’s retail cloud, it’s a ready-made bridge to the agentic future, avoiding costly custom integrations.
The timing is no accident. With the explosive growth of large language models and agent frameworks, retailers have been clamoring for a safe, standardized way to let AI assistants tap into core commerce operations. Microsoft’s answer is an MCP server that sits inside the Commerce Scale Unit, enforcing Entra ID authentication and granular permissions so only vetted agents can initiate real-world transactions.
What the MCP Server Actually Does
At its core, the MCP server exposes a set of structured capabilities – Microsoft calls them “tools” – that AI agents can invoke. These include:
- Product search and catalog browsing: retrieve product details, images, inventory levels, and pricing in real time.
- Cart management: add items, apply coupons, calculate totals with dynamic tax rules.
- Checkout orchestration: collect shipping information, validate addresses, and process payments via integrated gateways.
- Order life‑cycle: submit orders, retrieve order history, and handle returns or cancellations.
Each tool is wrapped in an OpenAPI-like specification that agent developers can inspect. This means an AI built for customer service – say, a shopping copilot on a retailer’s website – can now not only recommend a product but also place the order, all while respecting the store’s business logic and security boundaries.
Microsoft has baked the MCP endpoint directly into Commerce Scale Unit version 10.0.48, so retailers running that release (or newer) can enable it from the System Parameters page. No separate infrastructure is required. The server runs as part of the Scale Unit’s existing IIS host, listening on a dedicated port and accepting requests only after AAD token validation.
Security: Entra ID and Zero Trust by Default
The preview ships with mandatory Entra ID authentication. Every AI agent must obtain an access token – either via client credentials or delegated user flow – before it can touch the MCP endpoint. Inside the Commerce back office, administrators can grant scoped permissions to each agent, limiting which tools it can use and which channels or legal entities it can affect.
This zero‑trust model is intentional. “Retailers don’t want their AI bot accidentally ordering 10,000 units of something,” said a Microsoft program manager in a pre‑release webinar seen by Windows News. “Every call is logged, every action is auditable, and you can revoke access in real time.”
Audit trails appear in the standard Commerce activity log, showing which agent performed what action and on behalf of which user (if any). Microsoft Entra’s Conditional Access policies can further restrict requests to specific IP ranges or require multi‑factor authentication for high‑value operations.
How Developers Can Get Started
For partners and ISVs, the onboarding path is straightforward. After enabling the MCP server, an administrator generates a client app registration in Entra ID and notes the generated tool manifest URL. The AI agent’s runtime – whether it’s a custom Python microservice, an Azure OpenAI on‑your‑data setup, or a third‑party agent framework – fetches that manifest to discover the available tools and their parameters.
Microsoft has published a reference implementation on GitHub that shows how to connect an Azure AI Foundry agent to the MCP server. The sample uses the Semantic Kernel framework to chain a natural‑language prompt to the tool calls, confirming the cart and payment details before finalizing the order.
For quick testing, the Commerce Scale Unit includes a built‑in “MCP Explorer” page where administrators can invoke tools manually, inspecting the JSON requests and responses. This sandbox environment is ideal for debugging permission scopes or understanding the data contract before handing off to an AI model.
Real‑World Use Cases
Early adopters in the private preview phase have piloted scenarios that hint at the wider potential:
- Conversational commerce: A chatbot on a fashion retailer’s website uses the MCP server to pull live inventory from physical stores, suggesting alternatives when an item is out of stock, and completing the purchase without ever leaving the chat window.
- Voice‑assisted ordering: A drive‑thru quick‑service restaurant integrated the MCP endpoint with a voice agent, allowing customers to reorder their favorite meal by speaking their phone number – the agent matched the loyalty account, pulled the last order, and processed payment automatically.
- B2B sales copilot: A wholesale distributor’s field‑rep app queries the MCP server to create complex quotes with tiered pricing and contract‑specific discounts, dramatically reducing manual data entry.
- Fraud prevention: A custom AI model monitors transactions in real time; when it detects suspicious patterns, it calls the MCP server’s “hold order” tool to pause fulfilment until a human reviews the case.
These vignettes share a common thread: they turn the commerce engine into a programmable building block for intelligent workflows, rather than a monolithic system that only humans can operate.
Industry Reaction and Analyst Take
Retail technology analysts have greeted the preview with cautious optimism. “The MCP server is the missing link between generative AI and transactional retail systems,” noted Gartner VP Analyst Sarah Evans in a recent note. “While many vendors offer AI‑powered chat, few provide a secure, managed way to execute commerce operations. Microsoft’s approach lowers the risk for enterprises that want to experiment with agentic checkout.”
Community forums on platforms like Reddit and the Microsoft Dynamics community have seen lively discussion. Some retailers expressed concern about latency – each MCP call traverses the Scale Unit’s logic, which could add 200‑300 milliseconds compared to a direct database query. Microsoft’s documentation acknowledges the overhead and recommends caching non‑transactional data at the agent layer.
Others worry about cost. The MCP server does not carry an extra license fee during preview, but Microsoft has signaled that it will be part of the Dynamics 365 Commerce Premium licensing tier once generally available. “It’s a premium feature for a premium scenario,” one forum contributor wrote. “If you’re not building an AI agent, you won’t need it.”
The Road to General Availability
Microsoft has not announced a firm GA date, but sources close to the product team suggest a target of Q4 2026. Until then, the public preview is fully supported for testing and production‑use evaluation, though Microsoft’s standard preview SLA applies.
The roadmap also hints at future enhancements: support for asynchronous tool execution (useful for long‑running fulfillment processes), integration with Microsoft Copilot for Sales, and a plug‑in marketplace where ISVs can publish custom MCP tools that extend the server’s capabilities – for instance, connecting to a loyalty‑point engine or a 3rd‑party payment gateway.
Crucially, Microsoft plans to ship the MCP server as a built‑in component of all future Commerce Scale Unit releases, meaning retailers on the modern update cadence will get it automatically. This reflects a broader strategy of making the Dynamics 365 platform “agent‑ready” by default, not an add‑on afterthought.
What Retailers Should Do Now
For IT decision‑makers in retail, the public preview is a call to action. Even if a full‑blown AI agent isn’t on the immediate agenda, enabling the MCP server and exploring the tool manifest costs nothing and builds institutional knowledge. Here are three concrete steps to take today:
- Upgrade to Commerce Scale Unit 10.0.48 (or the latest release) and enable the MCP server in System Parameters > General > MCP Server.
- Register a test Entra ID application and grant it minimal permissions – for example, read‑only access to the product catalog. Use the MCP Explorer to verify that the agent can retrieve a product’s details.
- Experiment with the GitHub reference implementation to see how a Semantic Kernel agent translates a user’s natural‑language request into tool calls. This exercise will reveal gaps in your API‑centric architecture before those gaps become bottlenecks.
Microsoft is also offering a series of “MCP in a Day” workshops via its partner network, where retailers can pair with a certified solution integrator to build a proof‑of‑concept agent in a controlled environment.
A New Chapter for Retail AI
Today’s public preview is not the final destination but a definitive starting gun. The Dynamics 365 Commerce MCP server transforms the commerce engine from a passive record system into an active participant in the AI‑driven customer journey. It doesn’t replace the need for thoughtful UX design or robust business logic; it simply gives that logic a machine‑readable interface that today’s AI models can actually use.
For the Windows‑centric retail IT crowd, this release reinforces Microsoft’s commitment to keeping the Windows ecosystem at the center of enterprise innovation. The Commerce Scale Unit itself still runs on Windows Server (with containerized options on Kubernetes), and the MCP server’s IIS integration will feel familiar to any Windows admin. The agent wave doesn’t mean the end of the GUI; it means the GUI finally has a programmable sibling that can work at the speed and scale of AI.
The message to retailers is clear: the building blocks for autonomous commerce are here. How aggressively you assemble them is up to you.