Over the past few weeks, I’ve been diving deeper into what’s possible when you combine the Dynamics 365 MCP server with a multi‑agent architecture. My goal has been simple: can AI agents work together to configure a full Record‑to‑Report (R2R) process in Dynamics 365, similar to how a real ERP project team would?
My latest video walks through the experiment, the setup in VS Code, how the agents collaborate, and what actually happens inside D365 when everything runs. This blog post captures the highlights and gives a bit more background.
Why Agents for ERP?
ERP implementations follow well‑defined patterns. They involve predictable tasks, structured configuration steps, and clear hand‑offs between roles. That makes ERP a great playground for testing whether AI agents can coordinate the way human project teams do.
My first test used a single agent, and while it got surprisingly far, it didn’t feel realistic. ERP projects depend heavily on role separation, ownership, and decision flow.
So I rebuilt the entire experiment around a multi‑agent design; much closer to how actual projects operate.
The Multi‑Agent Setup in VS Code
Inside VS Code, I now have a set of agents that each handle their own responsibilities:
- Project Manager Agent – Orchestrates the process, delegates tasks, and manages context.
- Solution Architect Agent – Defines solution structure, setup decisions, and blueprint logic.
- Consultant Agent – Connects to the MCP server and performs the actual configuration steps.
- Data Migration Lead Agent – Loads opening balances and reference data.
- Tester Agent (coming soon) – Validates what was configured and generates Azure DevOps test cases.
This structure immediately changed the feel of the experiment. Instead of one agent doing everything, I now see agents passing decisions, coordinating actions, and contributing their part of the R2R process.
It’s still earl, but it’s starting to behave like a digital implementation team.
What’s Inside the Video
In the video, I walk through:
1. The Agent Architecture
How each agent is configured, what roles they play, and how the Project Manager coordinates everything.
2. My VS Code Environment
A tour of the folder structure, the skills library, the agent instructions, and the MCP server connection.
3. Running the R2R Scenario
You’ll see the agents:
- create a legal entity
- setup the fiscal calendar
- configure the chart of accounts
- produce a configuration report documenting their work
Not every step is perfect, but the system is clearly trending toward repeatability.
4. What Worked and What Didn’t
The experiment surfaces some quirks, like the data migration agent reporting a blockage even though the main accounts eventually showed up in the system. These edge cases highlight where instructions need refinement.
5. What’s Next
I share upcoming improvements including:
- better validation and error handling
- more specialized skills
- a Tester agent
- expanding beyond R2R into additional workstreams
Why I’m Sharing This Early
The goal isn’t perfection, it’s progress. Agentic ERP implementation is still new, and the only way we learn is by sharing and collaborating. I’m releasing this early version because I’m hoping it sparks conversation, ideas, and maybe new approaches from others experimenting in this space.
Want the Code?
If enough people are interested in seeing the full agent structure, instructions, skills, and VS Code setup, I’m happy to put everything into a GitHub repo.
Just comment on the LinkedIn post, blog, or video—if there’s interest, I’ll publish the code.
Leave a Reply