Dataverse Now Speaks Claude, Cursor, and Copilot: What Microsoft’s MCP Expansion Means for Dynamics 365 Teams

Dataverse Now Speaks Claude, Cursor, and Copilot: What Microsoft’s MCP Expansion Means for Dynamics 365 Teams

On July 6, 2026, Microsoft quietly closed a gap that has been slowing down every Dynamics 365 and Power Platform team that also wants to use Claude or Cursor: the Dataverse coding-agent plugin, previously limited to GitHub Copilot, is now available across the Claude and Cursor marketplaces too. If your developers have been maintaining separate, hand-rolled prompt libraries just to get an AI assistant to understand your Dataverse schema, this update is worth a close look.

Key Takeaways

  • Microsoft’s Dataverse coding-agent plugin, live for Claude and GitHub Copilot since April 30, 2026, added Cursor support on July 6, 2026, giving all three assistants the same Dataverse expertise through one plugin.
  • The plugin routes natural-language requests through 15 named, auditable MCP tools and inherits the calling user’s existing Dataverse security role — no new permissions model to manage.
  • Microsoft’s broader MCP catalog now lists more than 60 ready-to-use servers, alongside a new partner certification program and a “bring your own MCP” governance path for internal tools.
  • A separate but related feature, Business Skills, is now generally available: reusable, natural-language process definitions stored centrally in Dataverse that any MCP-connected agent can discover and follow.
  • The open-source microsoft/Dataverse-skills repository ships eight specialist skills (connect, query, data, metadata, solution, admin, security, overview) that install directly into Claude Code, Cursor, GitHub Copilot, or Codex.
  • For CRM/ERP teams standardizing on multiple AI coding tools, this removes the “which assistant knows our Dataverse environment” problem entirely.

What Actually Changed on July 6

Until this update, if your team wanted an AI coding assistant that genuinely understood a specific Dataverse environment — its entity model, its security roles, its solution structure — GitHub Copilot was effectively the only first-party option, alongside Claude as of an earlier April 30, 2026 rollout. Cursor users were left writing custom instructions from scratch or working blind.

Microsoft’s July 2026 Power Platform blog update describes the fix plainly: the plugin now provides “the same Dataverse expertise: intelligent skill routing, enterprise-grade guardrails” across all three surfaces. In practice, that means a developer can describe what they want — “add a new lookup field to the Opportunity table and update the related plugin,” for example — in Claude, Cursor, or Copilot, and the plugin decides whether to route that request through the Dataverse MCP server, the Python SDK, the PAC CLI, or the Dataverse CLI, without the developer needing to know which tool does what.

This is a meaningful shift from “AI assistant that can write generic C# or JavaScript” to “AI assistant that understands your specific Dataverse environment’s shape, rules, and security model.” That distinction matters enormously for CRM and ERP teams, where the cost of a well-intentioned but schema-blind AI suggestion is a broken plugin, a corrupted solution layer, or a security role that quietly grants too much access.

Why This Matters for Business Buyers, Not Just Developers

It’s tempting to file this under “developer tooling news” and move on. For organizations running Dynamics 365 Sales, Customer Service, or a custom Power Platform application, it’s more consequential than that for three reasons.

First, governance travels with the agent. The plugin enforces least-privilege access, performs server-side authorization, and inherits the calling user’s existing Dataverse security role — it doesn’t grant a new, broader identity to the AI. Microsoft’s documentation on the underlying microsoft/Dataverse-skills repository is explicit that the plugin “transmits only application metadata — not prompts or record data.” For any business that has been nervously watching AI coding assistants get closer to production customer data, that’s the detail that should reassure a security or compliance reviewer.

Second, the tool surface is now typed and auditable. A related June 8, 2026 update formalized what Microsoft calls the “tool shape” of the Dataverse MCP server — a fixed set of 15 named tools rather than an open-ended natural-language free-for-all. That’s a real distinction from earlier generations of AI-CRM integration, where “the AI can do anything you ask” was as much a liability as a feature. Fifteen named, auditable tools is a contract an IT department can actually review, log, and restrict.

Third, this is no longer just about Microsoft’s own Copilot. By extending first-party support to Claude and Cursor, Microsoft is implicitly acknowledging that enterprise developers are multi-model shops now — and that locking Dataverse expertise inside one vendor’s assistant doesn’t match how teams actually work.

Practical Use Cases We’re Watching

For consulting teams like ours who implement and extend Dynamics 365 and Power Platform solutions daily, a few concrete workflows stand out:

  • Schema-aware scaffolding. A developer working in Claude or Cursor can now ask for a new custom table with specific relationships and get output that actually matches the target environment’s naming conventions and existing entity structure, rather than a generic Dataverse-shaped guess.
  • Safer bulk data operations. Because the plugin routes through the Dataverse CLI and Python SDK under the calling user’s own role, a request like “deactivate all contacts that haven’t had activity in 18 months” is checked against that user’s actual permissions rather than executed with elevated, agent-level access.
  • Faster solution packaging and deployment. The dv-solution skill in Microsoft’s open-source skill set is built specifically for solution lifecycle management — export, import, versioning — a task that historically required deep familiarity with the PAC CLI.
  • Institutional-knowledge capture via Business Skills. Separately, but relevantly, Microsoft’s Business Skills feature reached general availability in the same July wave. These let a business analyst (not just a developer) document a specific process — approval steps, required fields, business rules — once in Dataverse, and have any MCP-connected agent, whether in Copilot Studio, VS Code, Azure AI Foundry, or a custom app, discover and follow it automatically. Update the skill once, and every agent that references it inherits the change.

That last point is worth sitting with. For years, “how we actually do things here” has lived in tribal knowledge, onboarding decks, and the heads of a handful of tenured employees. Business Skills gives that knowledge a durable, machine-readable home that AI agents can act on consistently — a genuine answer to the recurring complaint that AI agents are only as good as the institutional context nobody wrote down.

Benefits and Real Challenges

The benefits are straightforward: faster development cycles, less duplicated prompt-engineering effort across tools, a consistent security model regardless of which AI assistant a developer prefers, and a documented path for partners to certify their own MCP servers rather than relying on undocumented integrations.

The challenges are just as real. Rolling this out isn’t a flip-a-switch exercise. Microsoft’s own admin center guidance notes that, for existing Dataverse MCP server customers, only Microsoft Copilot Studio is enabled by default under the enhanced governance controls — meaning IT admins need to deliberately opt in and configure access for Claude, Cursor, or other MCP clients rather than assuming it’s already live. Organizations bringing their own internal MCP servers also need to work through Microsoft’s new certification and governance process rather than treating any MCP endpoint as automatically trustworthy. And teams that have already built custom prompt libraries or internal tooling to work around the absence of this plugin will need to evaluate whether to retire that work in favor of the officially supported path.

Implementation Best Practices

Based on how we approach every new agent-facing capability for clients, a few things matter most when adopting this:

  • Audit which security roles your developers use day to day before rolling out AI-assisted Dataverse access — since the plugin inherits the calling user’s role, an overly broad existing role becomes an overly broad AI capability.
  • Start with the read-heavy skills (dv-query, dv-metadata) before enabling write-capable ones (dv-data, dv-admin) for a new team, so developers build trust in the tool’s judgment before it can modify records or environments.
  • Document your top three or four recurring business processes as Business Skills first, rather than trying to capture your entire operating manual at once — the value compounds as agents actually use them, and a smaller, well-maintained set beats a large, stale one.
  • If you’re evaluating a partner-built or internal MCP server, confirm it has gone through Microsoft’s certification path (or your own governance review) before connecting it to a production Dataverse environment.
Coding AssistantDataverse Plugin AvailabilityInstall Method
GitHub CopilotSince April 30, 2026/plugin install dataverse@awesome-copilot
Claude CodeSince April 30, 2026/plugin install dataverse@claude-plugins-official
CursorAdded July 6, 2026/add-plugin dataverse or via Settings → Plugins
CodexSupportedAdd marketplace from the GitHub repository URL

CRM Experts Online’s Perspective

We implement and extend CRM and ERP systems for a living, and the pattern we see most often isn’t a lack of AI tooling — it’s fragmentation. One client’s developer swears by Cursor, another’s team is standardized on GitHub Copilot, a third is piloting Claude Code, and none of them have had a consistent, governed way to let those tools actually understand a specific Dataverse environment until now. That fragmentation has real cost: duplicated onboarding, inconsistent guardrails, and AI-generated changes that don’t respect the security model an admin spent months getting right.

What we like about this update is that it doesn’t ask a business to pick a single “winning” AI assistant. It asks Dataverse itself to be the source of truth — for schema, for security roles, for business process knowledge — and lets whichever coding tool a developer prefers plug into that same source. That’s the right architecture for how enterprise AI adoption is actually happening: multi-model, not single-vendor. When we scope a Power Platform or Dynamics 365 engagement now, part of our discovery is assessing whether a client’s existing security role structure is clean enough to safely hand to an AI-assisted developer workflow — because this plugin makes that structure matter more, not less. If your team is sitting on a pile of ad hoc AI prompts and workarounds for Dataverse development, this is a good moment to consolidate onto the supported path and have a governance conversation before scaling it further.

FAQ

Do we need a new Microsoft license to use the Dataverse plugin with Claude or Cursor? No new Dataverse license is required beyond having a Dataverse environment accessible through Power Apps, Dynamics 365, or the free Power Apps Developer Plan; you do need your own Claude or Cursor subscription for those tools themselves.

Does the AI assistant get broader access to our data than a developer already has? No. The plugin is built on least-privilege principles, performs server-side authorization, and inherits the calling user’s existing Dataverse security role rather than operating with elevated or separate permissions.

Is our record data sent to Claude, Cursor, or GitHub Copilot’s servers? According to Microsoft’s documentation for the underlying skills package, the plugin transmits only application metadata, not prompts or record data, when routing requests.

What is the difference between the Dataverse plugin and Business Skills? The plugin gives coding assistants structured access to build, query, and administer a Dataverse environment. Business Skills is a separate, complementary feature that lets business users document organizational processes in natural language so any MCP-connected agent can follow them consistently.

Is this only useful for custom development, or does it help standard Dynamics 365 Sales/Service usage too? It’s primarily a developer and admin tool for building, extending, and administering Dataverse environments — including those underlying Dynamics 365 Sales and Customer Service. It doesn’t change the end-user sales or service agent experience directly.

Do we have to enable this for every developer immediately? No. Microsoft’s governance controls default to only Microsoft Copilot Studio being enabled for existing Dataverse MCP server customers, meaning admins can and should deliberately opt in team by team.

What if we’ve already built our own AI tooling for Dataverse? It’s worth evaluating whether to migrate to the officially supported plugin and skill set, since it now carries Microsoft’s governance, certification, and security-role inheritance model rather than a custom, unaudited integration.

Conclusion

The headline is simple: Dataverse development is no longer tied to one AI coding assistant. The more important story underneath is that Microsoft is building a governed, auditable bridge between whichever AI tool your developers prefer and the security model, schema, and institutional knowledge already living in your Dataverse environment. For teams juggling Dynamics 365, Power Platform, and a mix of AI coding tools, that’s worth acting on deliberately rather than letting it happen by accident. If you’re planning a Dynamics 365 or Power Platform project and want help thinking through how AI-assisted development fits your security model and team’s tooling choices, schedule a consultation with our team.

Further Reading