Microsoft opens its Dataverse coding plugin to Claude, Cursor and Copilot

Microsoft has extended its Dataverse coding-agent plugin so it works across three of the most popular AI coding tools at once. In a July 6, 2026 update on the Power Platform blog, the company said the plugin now supports Claude, Cursor, and GitHub Copilot, so that — in Microsoft's words — "whether a developer's primary coding agent is GitHub Copilot, Claude or Cursor, they get the same Dataverse expertise."

Dataverse is Microsoft's managed data platform underpinning Power Platform and many enterprise apps. The plugin lets a coding agent build and manage Dataverse solutions from natural-language instructions: listing tables, inspecting schemas, composing queries, and proposing data changes without the developer hand-wiring each API call.

One plugin, many tools underneath

The headline idea is intelligent skill routing. Rather than exposing a single API, the plugin decides which underlying tool fits each request — the Dataverse MCP server for small reads and writes, the Python SDK for bulk operations, and the PAC CLI or Dataverse CLI for authentication and environment management. A developer asks for what they want; the plugin picks the path.

That routing is built on the Model Context Protocol (MCP), the open standard for connecting AI agents to external tools that has become common ground across vendors this year. Microsoft says its MCP catalog now includes 60-plus ready-to-use servers, with the Dataverse MCP natively supported in Copilot Studio and Azure AI Foundry.

Governance is the real story

For enterprises, the more consequential change is that the plugin does not bypass existing security. According to Microsoft's developer documentation, it enforces least-privilege access and respects existing Dataverse role-based access control (RBAC). It cannot grant itself permissions: a tenant admin must still approve the MCP client app once per tenant, and an environment administrator must explicitly allowlist that client before an agent can touch data. The plugin generates the consent URLs and walks users through the settings pages, but the human approvals stay in place.

The update also adds two governance features aimed at larger organizations. MCP certification lets independent software vendors package and certify partner-built MCP servers through Partner Center, and a bring-your-own-MCP path lets companies register internal tools and proprietary systems under the same enterprise controls.

Why it matters for people learning to build

The announcement is a clear sign of where AI-assisted development is heading: away from a single all-knowing chatbot and toward agents that plug into the specific systems a team already runs. The same MCP plumbing that connects Claude or Cursor to Dataverse also connects them to a database, an issue tracker, or a monitoring dashboard.

For learners, the practical takeaway is that "which coding agent should I pick?" matters less than it used to. As MCP support converges, the connection layer is increasingly shared — the skills you build understanding how agents reach external tools carry across Claude, Cursor, and Copilot alike. The governance details are worth noting too: even a fully autonomous agent still works within the permissions a human granted it, which is exactly how it should be.