It’s rare that a single feature in an application redefines an industry. Over the last couple of months I believe that the advent of the D365 ERP MCP server is the single most impactful feature that has been released on the application in a decade, of course there’s a little bit of help from modern AI advancement too.
The way we interact with ERP systems has been shifting—quietly at first, and now with some serious momentum. With the release of the new D365 ERP MCP feature in versions .46 and .47 of Dynamics 365 Finance & Supply Chain, our community has an opportunity to rethink how we configure, automate, and engage with the systems we work in every day.
This blog aims to give everyone—from consultants to customers—a clear, approachable walkthrough of what this feature is, why it matters, and how we can collectively benefit from it.
Context Deterministic vs. Stochastic Systems
Before diving deeper, it helps to set the scene.
ERP systems—especially D365 Finance & Supply Chain—are deterministic. Their behaviour follows rules, and outcomes are predictable. This is essential for accounting, regulatory compliance, and operational stability.
Stochastic systems, on the other hand, are probabilistic. Any solution that leverages a Large Language Model (LLM) falls into this category. Outputs may vary based on reasoning, randomness, and context. This is what makes MCP powerful—but it also introduces a unique set of considerations for organisations.
Introduction to the D365 ERP MCP Feature
Microsoft has introduced a significant new feature in D365 Finance & Supply Chain: ERP MCP (Microsoft Copilot Plugins). While still evolving, it opens a powerful new way of interacting with ERP systems using natural language and automation. This release has meaningful implications for our industry, not just from a technical perspective but also in how consultants and customers collaborate to solve problems.
The MCP (Model Context Protocol) was developed to allow a standard way for AI applications to interact with external systems. Microsoft has extended a MCP server that allows us to use AI applkications to now interact with D365 Finance and Supply Chain.
A setup guide can by found here by André Arnaud de Calavon.
Why the MCP Feature Matters
To really understand how we use the MCP server we first need to classify the different use cases into either production or non‑production is use-cases. The MCP unlocks incredible potential, but because it’s stochastic, organisations must think carefully about where and how it is used. It’s useful in both scenarios—but in very different ways.
The short version:
- Production = caution, governance, constraints
- Non-production = freedom, exploration, automation
This distinction becomes the thread tying together the rest of this discussion.
Security Considerations in Production based use Cases
When MCP interacts with D365, it typically operates using one of three types of toolsets:
- Form interactions
- OData CRUD operations
- Direct API/class calls
Each can read, write, update, and delete data—meaning production environments require extremely careful planning. MCP uses either the security profile of the user interacting with the agent or a dedicated service account, and both models carry risks if not controlled.
High privileges + stochastic behaviour = potential operational impact.
Because of this, Microsoft is exploring additional layers of security for production implementation, and until then, organisations need to approach MCP with caution in live systems.
Advantages and Practical Uses in Non‑Production
This is where MCP truly comes alive because we are less concerned about highly priveleged MCP interactions.
In non-production environments, we can safely unlock far more capability. We can let the MCP reason, explore, recommend, test, and learn—without jeopardising business‑critical systems. Common use cases already emerging in the field include:
- Automating build and configuration tasks
- Accelerating environment setup
- Performing advanced lookups
- Searching large volumes of data quickly
- Performing reasoning tasks.
One organisation recently used the MCP to perform high-volume data searches that were too heavy for D365’s interface. The potential here is enormous.
Automating Environment Setup & Configuration
One community member doing impressive work in this area is Fredrik Saetre, who has shared repositories demonstrating configuration agents using the D365 ERP MCP server. These examples show how MCP can ingest requirements, reason through them, build configuration plans, and execute them—saving countless hours during environment provisioning.
This kind of innovation represents a turning point for consultants, enabling us to focus more on design, quality, and strategic decisions rather than repetitive tasks.
Getting Started with MCP in Copilot Studio
Here’s the high-level process for getting everything up and running:
- Ensure your organisation holds the correct licenses.
- Enable the MCP feature in Feature Management in D365 (version 46/47 required).
- Configure required settings within the D365 application.
- Log into Copilot Studio, select your target environment, and create a new MCP-enabled agent.
Once set up, you can begin exploring the available actions and understanding the boundaries of what MCP can do.
More Advanced Use Cases: VS Code & GitHub Copilot
Beyond Copilot Studio, you can interact with multiple MCP servers—such as D365, Azure DevOps, and Microsoft Learn—directly from VS Code using GitHub Copilot.
This combination allows MCP to:
- Parse and understand documentation
- Build structured requirement models
- Generate runtime plans
- Execute configuration tasks
- Update DevOps for traceability
This reduces the “black box” problem, making stochastic systems behave more predictably through structured references and orchestration patterns.
Reasoning Models & Best Practices
The meeting also provided valuable guidance on choosing the right model for each task:
- Opus 4.6 → best for deep reasoning
- Sonnet 4.6 → strong all-rounder
- Haiku → ideal for lightweight tasks when detailed prompts are provided
But the golden rule remains: good inputs create good outcomes.
Poor instructions cause confusion, context rot, and unpredictable behaviour. Clear, structured prompt design remains crucial.
Future Outlook
We expect to see faster implementations, reduced configuration time, and more empowered business teams as MCP matures. The ERP itself will not disappear—its deterministic foundation remains essential—but the way we interact with it is already transforming.
For the first time, we’re moving beyond strict reliance on graphical interfaces and toward more intelligent, natural, and flexible interactions. The community’s early experimentation will shape what this future looks like.
In Closing
Thank you for taking the time to dive into the MCP landscape with me. This technology represents an exciting shift, and its success will be driven not just by Microsoft’s advancements but by our community’s shared learning and experimentation.
If you have questions, want to collaborate, or wish to share your own experiences, I’d love to hear from you. Please feel free to comment, subscribe, or reach out anytime.
Author: Brendon Breedt
Date 26/02/2026
Important persons
Referenced persons in the article
Brendon Breedt
Position or title
