Microsoft now ships three separate MCP servers for the Dynamics 365 CRM ecosystem: one for Dataverse, one for Dynamics 365 Sales, and one for Dynamics 365 Customer Service, which reached general availability on 30 July 2026. Model Context Protocol, MCP, is the open standard that lets an AI agent, whether that’s a Copilot Studio agent, GitHub Copilot in VS Code, or another MCP-compatible client, discover and call tools against your live CRM data instead of you hand-building a custom connector for every scenario.
This guide covers what each of the three servers actually does, how to connect one to a Copilot Studio agent or to VS Code, what you can ask it once it’s connected, and the licensing and support limitations worth knowing before you build on it.
What MCP Actually Changes for Dynamics 365
Before MCP, getting an AI agent to read or write Dynamics 365 data meant building a custom connector or a Power Automate flow for each specific action, the pattern we cover in our guide to custom actions in Copilot Studio. MCP servers work differently: they expose a set of general-purpose tools, list tables, describe a table’s schema, read records, run a query, and any MCP-compatible agent can discover and call those tools directly, without you wiring up a bespoke action for every question someone might ask.
The practical effect is that an agent can answer something like “how many accounts do I have” or “describe the account table” without a developer having pre-built that exact capability. That flexibility is also why MCP tools are usually layered on top of the deterministic and hybrid control boundaries we set out in our piece on generative orchestration architecture, since a general-purpose query tool needs the same care around what it’s allowed to touch as any other agent capability.
Three Servers, Not One
“The Dynamics 365 CRM MCP server” is really three distinct servers, each covering different ground, and most real projects end up connecting more than one.
| Server | What it’s for | Status |
|---|---|---|
| Dataverse MCP Server | Schema exploration and raw CRUD: list tables, describe a table, read and write records, run queries | Available in Copilot Studio and via VS Code |
| Dynamics 365 Sales MCP Server | Sales-specific agent tools: qualification agent and opportunity research capabilities, built for sales scenarios rather than raw data access | Available; connects alongside Dataverse MCP for CRUD operations |
| Dynamics 365 Customer Service MCP Server | Customer Service agent tools and case data | Reached general availability 30 July 2026 |
If you only need an agent to answer questions about CRM records, Dataverse MCP alone is often enough. If you want an agent to actively work sales opportunities, Microsoft’s guidance is to connect the Sales MCP server for the sales-specific tools and the Dataverse MCP server alongside it for any CRUD operations the Sales tools don’t cover.
Connecting the Dataverse MCP Server in Copilot Studio
The Dataverse MCP server is built into Copilot Studio as a tool you add to an existing agent, not a separate integration to configure from scratch:
- In Power Apps, select your environment, then go to Agents and open or create the agent you want to connect it to.
- Scroll to the Tools section and select + Add tool.
- Select Model Context Protocol, then Dataverse MCP Server. If you don’t already have a Dataverse connection, you’ll be prompted to create one.
- Select Add to agent.
That’s the whole setup. Test it in the agent’s test pane with something like “show me the tables in Dataverse” or “how many accounts do I have,” and the agent calls the MCP tools directly rather than needing a topic built for that specific question. One prerequisite worth checking first: your environment admin needs to have the Copilot Studio MCP client allowed for the environment, which is an environment-level setting, not something you configure per agent.
Connecting From VS Code With GitHub Copilot
Developers can also reach these servers directly from VS Code, which is useful for testing tool behaviour before wiring an agent up in Copilot Studio. Open the command palette, run MCP: Add Server, choose the HTTP server type, and point it at the right endpoint. Dataverse uses https://<YourOrgURL>/api/mcp. The Sales MCP server uses a different pattern entirely, routed through the Agent 365 platform: https://agent365.svc.cloud.microsoft/mcp/environments/<EnvironmentID>/servers/msdyn_SalesMCPServer.
Once both servers show a working Start button in your mcp.json configuration, open the GitHub Copilot chat pane and prompt it directly, for example “list the leads that were updated since my last login.” The agent resolves that through the Sales MCP tools rather than a hand-built query.
What You Can Actually Ask It
The Dataverse MCP server’s tools cover schema exploration, listing tables and describing a table’s structure, alongside reading, creating, and updating records and running arbitrary queries. In practice that supports questions like “describe the account table,” “show me my contacts,” or “how many open opportunities does the Malaga team have.” The Sales MCP server layers qualification and opportunity-research tools on top, so it can support more sales-specific requests like summarising an account or identifying which leads need follow-up, using tools purpose-built for that scenario rather than a raw query.
Licensing, Credits, and a Real Limitation
Using the Sales MCP server’s tools consumes Copilot Studio credits, at a rate that varies by tool, so it’s worth checking Microsoft’s current consumption rates before assuming a rollout is free once the licences are in place. You’ll also need admin permissions in both Dynamics 365 Sales and Copilot Studio to set the connection up in the first place.
One limitation worth knowing before you plan a project around this: as of this guide, Claude Desktop is explicitly not supported as an MCP client for the Dynamics 365 Sales MCP server, even though MCP is an open, cross-vendor standard. Copilot Studio and GitHub Copilot in VS Code are the documented, supported clients today. If a client’s MCP support matters to your project, confirm it against Microsoft’s current documentation rather than assuming compatibility because the protocol itself is open.
Security and Governance
MCP access authenticates through OAuth 2.0 and respects your existing Dataverse security roles, so an agent connected via MCP can’t see or change more than the connecting user’s permissions already allow. That’s a meaningful guardrail, but it’s still worth deciding deliberately which of an MCP-connected agent’s actions should run freely and which need a confirmation step, the same control-layer thinking we cover in our generative orchestration architecture guide. A read-only query tool is low-risk; giving an agent unrestricted write access to production records is a different decision entirely, regardless of what the security role technically permits.
Frequently Asked Questions
Do I need a custom connector if I’m using the Dataverse MCP server?
Not for general schema exploration or CRUD operations, the MCP server already covers that. Custom actions and connectors still earn their place for calling external, non-Dataverse systems, or for exposing a very specific business process as a single named tool rather than a general query capability.
Can I use the MCP server with an AI assistant outside the Microsoft ecosystem?
MCP itself is an open, cross-vendor standard, so in principle any compliant client can connect. In practice, Microsoft’s current documentation names Copilot Studio and GitHub Copilot in VS Code as supported clients for the Dynamics 365 Sales MCP server, and explicitly states Claude Desktop isn’t supported at this time. Check current documentation before committing a project to a specific client.
Is the Dataverse MCP server free to use?
The server itself is a built-in Copilot Studio and Power Platform capability, but tool calls through it, and especially through the Sales MCP server, draw on Copilot Studio credit consumption, which varies by tool. Budget for this the same way you would budget for any other Copilot Studio usage, rather than assuming it’s included at no additional cost.
Does this replace Copilot Studio’s existing Dataverse connector?
No, they coexist. The Dataverse connector is called explicitly from within a topic. The Dataverse MCP server exposes its tools directly to the planner in generative orchestration, so the agent can choose to use them without you scripting the exact question in advance. Which one fits depends on whether you need a predictable, scripted action or a flexible query capability.
Next Steps
If you’re already running Copilot Studio agents against Dynamics 365, adding the Dataverse MCP server is a low-effort way to test whether general-purpose query access covers more of your use cases than the custom actions you’ve already built. AlishBit implements Dynamics 365 CRM and Copilot Studio agents for businesses across Spain and Europe, including MCP-based integrations, and we’re happy to review whether MCP fits your project.
Building AI Agents on Dynamics 365?
AlishBit implements Dynamics 365 CRM and Copilot Studio agents for businesses across Spain and Europe, including MCP server integrations. Book a free consultation and talk through your project.