Bring Your Own AI: What Dynamics 365’s New Openness Means for Business Leaders
Table of Contents
Dynamics 365’s Model Context Protocol (MCP) is an open standard, introduced at Microsoft Build, that lets any AI model, not just Copilot, connect securely to Dynamics 365 data and business logic through a single standardized interface. The Dynamics 365 ERP MCP server is in public preview, and MCP support in Copilot Studio is now generally available. Basic MCP-enabled Copilot features are included in existing Dynamics 365 and Microsoft 365 Copilot licenses; advanced agent usage is billed against Copilot Credits, priced separately.
Introduction
For years, Microsoft’s enterprise AI strategy was simple: use Copilot, or build a custom integration yourself. Every specialized AI model meant a new point-to-point connection, and every major Dynamics update put that connection at risk of breaking.
With MCP, that equation is changing. This guide breaks down what MCP actually does, what it costs in 2026, where it delivers real business value, and what to watch for before you build on it.
What is Model Context Protocol (MCP) in Dynamics 365?
MCP is a standardized connection layer between Dynamics 365 and any AI model.
Instead of custom point-to-point integrations for each AI tool, a single protocol handles data access, permissions, context, and audit logging.
Microsoft introduced Dynamics 365 MCP servers at Build 2025 for both ERP and CRM (CE) workloads. The specification is open; any AI vendor can build support for it without licensing it from Microsoft. In practice, MCP standardizes four things:
- Data access and permissions: What the AI is allowed to query
- Context management: What information does the AI receive about a request
- Security boundaries: Role-based access carried over from Dynamics
- Audit trails: A record of what was asked, by whom, and when
How Does Dynamics 365 MCP Actually Work?
An MCP server sits between Dynamics 365 and the AI model, enforcing permissions before any data reaches the model. The flow works in five steps:
- A user request enters Dynamics 365 with the user’s security context attached.
- The MCP server validates permissions and assembles the relevant data context.
- The AI model receives only what the user is authorized to see.
- The response passes back through MCP for logging and validation.
- The audit trail captures the interaction for compliance review.
This makes the protocol model-agnostic. GPT-based models, Claude, an industry-specific model, or an in-house model can all connect the same way; the integration method doesn’t change based on which AI is on the other end.
You may also like: Enterprise MCP Integration: Build, Test and Govern
What Does Dynamics 365 MCP Cost in 2026?
MCP itself isn’t a separate line item; it’s a connection protocol, not a licensed product. What you pay for is the Copilot and Dynamics 365 licensing around it, and two changes from the past year matter most for MCP-connected AI initiatives specifically:
- Basic MCP-enabled Copilot features are now bundled into existing Dynamics 365 app licenses and the standard Microsoft 365 Copilot license; the standalone Copilot for Sales SKU, for example, has been retired and rolled into the standard license.
- Advanced, agent-driven MCP workflows consume Copilot Credits rather than a flat per-user fee, so cost planning should be based on expected agent interaction volume, not just headcount.
For the full current breakdown of Copilot license tiers, Copilot Studio credit pricing, and Dynamics 365 licensing options, see our Dynamics 365 Licensing Guide and Guide to Microsoft Copilot Pricing & Licensing.
Ready to Evaluate MCP for Your Dynamics 365 Environment?
AlphaBOLD helps enterprise Dynamics 365 teams assess whether MCP fits their AI roadmap and architect the integration around real business requirements.
Request a ConsultationWhy Does MCP Matter for Enterprise AI Strategy?
The traditional integration pattern was slow: build a custom connection to a specialized AI model, watch it break on the next major update, then rebuild. MCP replaces that with a standard protocol that AI vendors build once and reuse across any Dynamics instance.
The strategic upside is flexibility. Organizations can swap AI models without rebuilding their integration layer, and security enforcement occurs at the protocol level. Role-based access from Dynamics carries through to whatever AI is connected, so the model only ever sees what the user is permitted to see.
- No repeated rebuild cycles when an AI vendor pushes a major update
- Model portability switch or add AI models without re-architecting the integration
- Protocol-level security. Dynamics permissions carry through automatically, rather than being re-implemented per model
- Reduced single-vendor risk for long-term AI roadmaps
You may also like: Generative AI with MCP for Seamless ERP & CRM Integration
Where Does MCP Deliver Practical Value?
MCP proves most useful in scenarios where a generic, one-size-fits-all AI model falls short of specific business, compliance, or technical requirements. Three patterns show up most often in enterprise deployments today.
- Regulated Industries: A healthcare organization that can’t risk a generic model hallucinating medical guidance can connect a healthcare-specific model through MCP. It inherits Dynamics’ existing permission structure and processes only sanitized data; there is no separate compliance layer to build.
- Global Operations: A multinational running English-centric AI across all regions often gets poor results due to local regulations and business practices. MCP allows different models per region, routed based on user location, while data governance remains centralized in Dynamics.
- Specialized Workloads: A manufacturer with a proprietary failure-prediction model trained outside Dynamics can connect it directly to field service workflows via MCP, without rebuilding the model to fit Microsoft’s framework.
What Are the Challenges of Adopting MCP?
MCP lowers the integration barrier, but it doesn’t remove governance work. Three areas need attention before scaling:
- Governance complexity. Multiple connected models mean multiple things to vet, approve, and monitor for behavior drift.
- Configuration effort. MCP standardizes the connection, but tuning a model to use Dynamics data well still requires understanding your schema and context setup.
- Vendor overhead. Each additional AI model brings its own SLA, pricing structure, and update cycle to manage.
For organizations with straightforward needs, Copilot’s built-in functionality may still be the simpler path; not every use case justifies running multiple models.
Who Should Prioritize MCP Adoption?
MCP adoption pays off fastest for organizations already running into the limits of a single, generic AI model, not for every Dynamics 365 customer. The clearest fits share one of a few traits.
MCP is most relevant for:
- Regulated industries where general-purpose AI models fall short of compliance requirements
- Global organizations that need region-specific AI capabilities
- Companies with proprietary AI/ML want to be connected to Dynamics 365
- IT leaders are actively trying to reduce single-vendor AI lock-in.
What Is Microsoft's Strategy Behind Opening Up Dynamics 365?
Microsoft isn’t stepping back from Copilot; it’s positioning on Dynamics 365 as the platform layer that works with any AI, on the logic that Microsoft still wins on infrastructure even when a customer chooses a non-Microsoft model.
For enterprises, that reduces the risk of tying an entire AI roadmap to one vendor’s release schedule.
- Platform over product: Microsoft profits from being the infrastructure layer, not just from Copilot licenses
- Value even on “losses”: Dynamics still captures value when a customer picks a non-Microsoft model
- Lower switching risk: less exposure to any single vendor’s AI roadmap or release cadence
- Dynamics as the system of record: AI models connect to it, rather than it being one AI product among many
You may also like: MCP Servers Explained: What They Are and How They Work
Not Sure Which AI Models Belong in Your Dynamics 365 Stack?
We'll help you map your use cases to the right mix of Copilot and MCP-connected models, without over-building or overspending.
Talk to an ExpertConclusion
MCP doesn’t force a choice between Copilot and everything else; it removes the need to choose at all. Organizations can keep Copilot for general-purpose work and connect specialized, regional, or proprietary models where they add more value, all through the same standardized, permission-aware connection.
The technology is still early: the ERP MCP server is in public preview, and best practices for governance and model selection are still forming. But the direction is clear enough to act on now. For IT leaders evaluating their next 12–18 months of AI investment in Dynamics 365, MCP is less a feature to enable than a shift in how much flexibility that investment can retain.
FAQs
Yes. MCP can run alongside existing integrations. Organizations can adopt it gradually as they replace or update custom connections.
You can connect another MCP-compatible model without rebuilding the Dynamics-side integration, since MCP uses a standard connection method.
No, but governance is recommended before connecting multiple models. Define model approval, access, and monitoring rules before expanding AI connections.






