Dynamics 365 Commerce · Agentic Commerce
D365 Commerce MCP Server: the beginning of agentic commerce
Microsoft's Commerce MCP server lets AI agents call the same catalog, pricing and inventory logic your storefront already uses. No replatforming required — but the governance questions arrive immediately.
The agentic path
Traditional commerce follows a familiar path: search, product, cart, checkout, order. Agentic commerce reshapes that path around customer intent instead.
At NRF 2026, Microsoft introduced the Dynamics 365 Commerce MCP server — a preview capability that connects AI agents to the Commerce engine retailers already operate, using the open Model Context Protocol (MCP). No new commerce platform. No replatforming. Just a standardized way for AI agents to call the capabilities that already exist.
01What is Model Context Protocol?
MCP is an open standard that lets AI agents discover and call the capabilities exposed by a backend system, instead of every AI application needing a custom-built integration. An MCP server exposes a defined set of tools — discrete actions like “search products” or “get order status” — that any compatible MCP client can call in a consistent way.
02What the D365 Commerce MCP server does
The Commerce MCP server exposes the Dynamics 365 Commerce headless engine — the same Commerce Scale Unit (CSU) that already powers e-commerce, in-store, call center, and mobile channels — to AI agents. The agent doesn't get a separate commerce stack; it calls the same catalog, pricing, inventory, promotions, and order logic every other channel uses.
- No replatforming — pricing, inventory, and promotions stay in one place.
- Omnichannel consistency — a discount that applies on the website applies the same way through an agent.
- Agent-platform neutrality — works with Copilot Studio, ChatGPT, Azure AI Foundry, or a custom agent host.
Commerce MCP is a preview capability, available on cloud-hosted Commerce Scale Units running version 10.0.48 or later. It is not supported on self-hosted or on-premises deployments, and preview terms are subject to change.
03What shoppers can actually do
A shopper might tell an AI agent: “Find a waterproof hiking jacket under $150 that I can pick up tomorrow near me.” The agent searches the catalog, checks nearby inventory, applies eligible promotions through the existing Commerce pricing engine, builds a cart, retrieves delivery options, and initiates a secure, hosted checkout. Dynamics 365 Commerce remains the system of record throughout — the agent orchestrates, but never calculates a price or invents a stock number on its own.
| Capability | Example |
|---|---|
| Product discovery | “Find running shoes under $100” |
| Inventory & availability | “Is size 8 available near me?” |
| Cart & checkout | “Add this to my cart and check out” |
| Promotions & discounts | “Apply the best discount I qualify for” |
| Order status & history | “Where is my order?” |
04AI-assisted commerce, not autonomous commerce
It's important to be precise about what this capability is — and isn't. Commerce MCP enables AI-assisted commerce: an agent that can look up products, build a cart, and initiate checkout within a permissioned, auditable framework, typically with the shopper completing payment through a secure hosted link.
That is meaningfully different from fully autonomous commerce, where an agent independently spends a customer's money with no human in the loop. Design your agentic commerce experiences with that distinction — and the right checkpoints — clearly in mind.
05The real question: what should you expose?
The real question isn't whether AI can connect to Commerce — it's which Commerce capabilities should be safely exposed to AI agents, and under what controls. Commerce MCP authentication is anchored in Microsoft Entra ID: agents call the server under a user context with a bearer token, and the Commerce platform applies channel, role, and customer-scoped permissions to every tool call.
Before rolling this out, work through:
- Security & authentication — how the MCP endpoint and agent identities are controlled
- Governance — which agent platforms are explicitly allowed to connect
- Business rules — keeping pricing and promotion logic owned by Commerce, not duplicated
- Human oversight — which actions proceed automatically vs. need a confirmation step
- Transaction controls — audit logging and accountability for every agent-assisted order
The guiding principle: expose capabilities deliberately, not indiscriminately.
06Why this matters for D365 Commerce customers
For organizations already running Dynamics 365 Commerce, this is additive, not disruptive: new AI-driven interaction channels, without duplicating pricing, inventory, or promotion logic, and without sacrificing the omnichannel consistency customers expect between agent, web, store, and call center.
07Final takeaway
The future of commerce may not be another storefront — it may be an intelligent agent that understands customer intent and interacts with the commerce engine on the customer's behalf. With MCP, Dynamics 365 Commerce is opening another path between AI experiences and enterprise commerce capabilities. The organizations that treat security, governance, and human oversight as first-class design questions will be the ones best positioned to make that path work.
Thinking about bringing AI into your D365 Commerce ecosystem?
Rolling out agentic commerce touches Commerce architecture, identity, security, integration, and governance all at once. Dexnen works across Commerce architecture, Commerce Scale Unit implementation, Azure integration, identity and security design, and ongoing managed services.
Talk to DexnenSources & further reading
- Microsoft Learn — Dynamics 365 Commerce Model Context Protocol (MCP) server (preview)
- Microsoft Dynamics 365 Blog — Dynamics 365 Commerce introduces agentic capabilities with MCP
- Anthropic — Introducing the Model Context Protocol
The Dynamics 365 Commerce MCP server is a preview capability at the time of writing. Preview functionality is subject to change — confirm current status on Microsoft Learn before making implementation decisions.