Building an AI Agent for Business Central Using Copilot Studio & MCP

Building an AI Agent for Business Central Using Copilot Studio & MCP

AI agents are moving fast from experimentation to real enterprise usage. When AI starts interacting directly with ERP systems, security, permissions, and governance become non-negotiable. In this INFOC-hosted YouTube session, we demonstrate how to build a secure, permission-aware AI agent for Microsoft Dynamics 365 Business Central using Microsoft Copilot Studio and the Model Context Protocol (MCP), which is currently in preview. This approach enables AI agents to query and act on Business Central data without custom connectors, hardcoded logic, or bypassing ERP security.

Why MCP Changes the Game for Business Central AI

Traditional AI integrations with ERP systems usually require:

• Custom APIs or connectors
• Middleware layers
• Hardcoded business logic
• Duplicated permission models

MCP eliminates this complexity. With MCP:

• Business Central exposes ERP APIs as tools
• Permissions are enforced natively by Business Central
• AI agents operate strictly within ERP security boundaries
• No custom connector development is required

This creates a clean, scalable, and governance-ready foundation for AI agents.

Step 1: MCP Setup in Business Central

The session begins by enabling and activating the MCP Server in Business Central.

Key configuration steps include:

• Enabling the MCP server
• Activating MCP at the environment and company level
• Defining which Business Central APIs are exposed as tools

For safety and governance, the demo starts with: Read-only API exposure: ideal for finance, reporting, and operational insights. Optionally, MCP also supports:

• Write-enabled actions (create, modify, delete)
• Controlled through Business Central permissions and governance rules

This ensures AI agents can never exceed the access rights of the ERP user context.

Step 2: Creating the AI Agent in Copilot Studio

Next, the AI agent is created using Microsoft Copilot Studio.

The setup includes:

• Creating a new agent
• Defining agent instructions and behavior
• Selecting the appropriate AI model
• Connecting the agent to the Business Central MCP server

At this stage, the agent becomes ERP-aware while remaining fully governed by:

• Business Central permission sets
• MCP-defined tool exposure
• Organisational access policies

No shortcuts. No backdoors.

Step 3: Dynamic Tool Mode: Solving the 70-Tool Limitation

A key highlight of the session is Dynamic Tool Mode. By default, AI agents face limitations on the number of tools they can work with. Dynamic Tool Mode:

• Allows tools to be discovered dynamically at runtime
• Removes the static 70-tool limitation
• Makes MCP-based ERP agents scalable and future-proof

This is especially important for Business Central environments with multiple modules and growing API exposure.

Step 4: Testing with Real Business Central Queries

The demo then moves into real-world testing with a natural language query: “What is the accounts receivable for School of Fine Arts?” The AI agent:

• Uses MCP tools to query Business Central
• Intelligently resolves fuzzy customer names
• Identifies the correct customer record
• Retrieves open invoices and A/R balances
• Drills into transactional details

This demonstrates how MCP enables context-aware ERP intelligence, not just surface-level chatbot responses.

Governance, Safety, and Enterprise Readiness

Throughout the session, one principle remains constant: AI agents must be safe by design. This architecture ensures:

• ERP permissions are always enforced
• Read and write actions are explicitly controlled
• No hardcoded logic bypasses Business Central security
• Full auditability and traceability

This aligns strongly with modern enterprise AI governance practices:

• Least-privilege access
• Human accountability
• Controlled autonomy
• Production-grade safety

INFOC Perspective

At INFOC, we see this pattern as the future of ERP-connected AI:

• Business Central remains the system of record
• MCP becomes the secure interface
• Copilot Studio acts as the agent orchestration layer
• AI delivers insights and automation without compromising control

This is how AI moves from demo environments to real business impact.

Watch the Full INFOC YouTube Session

Building an AI Agent for Business Central Using Copilot Studio & MCP This session is ideal for:

• Business Central consultants and architects
• ERP and IT leaders
• AI and automation teams
• Organisations exploring governed AI agents

Final Thoughts

AI agents are no longer optional experiments. They are becoming core enterprise capabilities. But success depends on how they are built. By combining Business Central, MCP, and Copilot Studio with strong governance, enterprises can unlock AI-driven ERP intelligence without sacrificing security or control. That is the difference between AI hype and AI that actually works.