On July 6, 2026, Microsoft announced that Dataverse is breaking out of the Power Platform bubble. The company is bringing its coding-agent plugin to three heavyweights of AI-assisted development: Anthropic’s Claude, the Cursor editor, and GitHub Copilot. The expansion means developers can now tap directly into Dataverse data from their favorite coding tools, with enterprise-grade governance layered on through the Model Context Protocol (MCP).
For the millions of businesses that store customer records, inventory data, and line-of-business tables in Dataverse, the move transforms how professional developers and citizen makers interact with that data. No more context-switching between low-code editors and pro-code IDEs—the data now surfaces right where code is written.
What actually changed: plugin reach and MCP controls
Until now, Dataverse’s coding-agent plugin was confined to Microsoft’s own Copilot experiences inside Power Apps and Power Automate. The July update opens it up to three external platforms that have become dominant in everyday development work:
- Claude (via MCP servers): Anthropic’s AI can now connect to a Dataverse MCP endpoint, allowing Claude to list tables, inspect schemas, compose FetchXML queries, and even propose data updates—all within a conversation thread.
- Cursor: The popular AI-native code editor gains a new Dataverse extension that surfaces real-time schema information and sample data as you type. It also honors MCP governance policies, so the agent only sees what it’s allowed to see.
- GitHub Copilot: Already deeply integrated with Azure and GitHub, Copilot now understands Dataverse as a first-class data source. When a developer prompts "show me the account table fields," Copilot can reach into the organization’s Dataverse environment and return accurate metadata.
All three integrations share a common governance backbone: MCP-based access control. The Model Context Protocol, originally proposed by Anthropic and adopted across the industry, standardizes how AI applications request and receive context from external systems. Microsoft has extended its MCP support in Dataverse to include:
- Azure AD authentication: Every agent request must present a valid token. Service principals and delegated user tokens are both supported, so background automation and interactive sessions are equally controlled.
- Row-level and column-level security: Dataverse’s existing security roles and field-level permissions are enforced during MCP interactions. An agent running under a user’s identity will only see records that user is allowed to access.
- Audit logging: All MCP requests are written to the Microsoft Purview compliance portal. Administrators can track which agent asked for what data, when, and from which IP address.
- Rate limiting and threshold alerts: To prevent runaway agent loops, organizations can cap the number of MCP requests per minute and receive alerts when limits are approached.
According to a Microsoft technical advisory reviewed by windowsnews.ai, the Dataverse plugin for Claude and Cursor ships as a pre-configured MCP server manifest—essentially a JSON file that points to the Dataverse endpoint and declares the available tools. For Copilot, the integration is built into the GitHub Copilot Chat extension, with a new checkbox in Visual Studio Code’s settings: “Enable Dataverse context.”
What it means for you
The impact splits along two fault lines: the developer writing code and the administrator guarding the data.
If you’re a developer
You can now stay inside your IDE and ask natural-language questions about live Dataverse structures. For example, in Cursor, typing // fetch all open cases from Dataverse can produce a ready-to-run Web API call, complete with the correct entity set name and filter syntax derived from real metadata. In Claude, you can upload a specification and ask the model to generate a Canvas App formula that references the actual column names of your custom table.
More importantly, the integration reduces the friction of building .NET or JavaScript back-ends that consume Dataverse. Copilot can scaffold a complete HttpClient method with the right $select and $filter because it has live access to attribute types and relationship names. Early adopters in the Power CAT community report that writing Dataverse-connected code feels as natural as querying a local database—without the need to maintain stale mock schemas.
If you are in IT or admin
Governance is the headline feature. The MCP controls allow you to treat AI agents like any other data consumer in your tenant. You can create a dedicated app registration for each coding tool, grant minimal privileges, and monitor usage through the Dataverse analytics dashboard. If an agent starts hammering the endpoint, threshold alerts notify you before the environment’s API call limits are exhausted.
A common worry among compliance teams—that AI agents might sidestep security roles—is addressed by the strict enforcement of Dataverse’s model-driven security model. An agent running as a user who only has read access to the contact table cannot modify it, even if the developer’s prompt says “update all contacts.” The MCP layer simply rejects the operation and logs the attempt.
How we got here
The announcement is the latest milestone in a three-year push to make Dataverse a universal agent data platform. In late 2025, Microsoft first added MCP endpoints to Dataverse, initially for internal Copilot experiments within Power Apps. That early implementation revealed that developers wanted to bring their own AI assistants, not just the ones Microsoft supplied.
At the same time, coding agents evolved from novelty to necessity. Cursor passed one million active users in March 2026, and Claude 3.5 Opus became a favorite among backend developers for its schema-aware coding abilities. GitHub Copilot, meanwhile, kept deepening its knowledge base—first with Azure services, then with third-party APIs, and now with organizational data locked in Dataverse.
The industry-wide adoption of MCP provided the missing transport layer. Instead of building a custom plugin for each tool, Microsoft instrumented Dataverse as an MCP server that any compliant client can consume. That elegance is evident in the announcement: the same governance policy written once applies to Claude, Cursor, and Copilot uniformly.
What to do now
If your organization uses Dataverse and any of the supported coding agents, here is how to turn on the connection:
- Enable MCP in Dataverse: Navigate to the Power Platform admin center, select your environment, and toggle on “Model Context Protocol endpoints.” This setting is under Settings > Integrations > AI tools. It typically takes a few minutes to propagate.
- Set up authentication: Create an app registration in Azure AD for each agent you plan to allow (or use a single multi-tenant registration with strict redirect URIs). Grant the registration the “Dataverse User” or custom security role. For user-delegated flows, no extra registration is needed—the agent simply passes the user’s OAuth token.
- Install the plugin
- Claude: Download the MCP server manifest from the Power Platform admin center (Integrations > MCP > Download manifest). In Claude, go to Plugins, select “Install custom MCP server,” and upload the manifest.
- Cursor: Open the Extensions marketplace, search “Dataverse (Preview),” and install. Restart Cursor and sign in with your organizational account.
- Copilot: In VS Code, open Settings, search “github.copilot.chat.dataverse.enabled,” and check the box. You may need to reload the window. - Test gradually: Start with a non-production environment. Run simple queries like “list all tables” to confirm the agent sees the expected entities. Then increase complexity, watching the Azure AD sign-in logs and the Dataverse API call volume.
- Set alerts: In the Power Platform admin center, go to Analytics > Data Events, and configure a threshold alert for MCP requests. A recommended starting point is 1,000 requests per hour per agent, adjusting based on your normal workload.
Microsoft’s documentation includes step-by-step walkthroughs for each code editor. Links will be added here as they become available.
What’s next
This release is labeled “public preview,” but Microsoft has signaled that general availability is targeted for Ignite 2026 in November. Between now and then, expect rapid iteration: support for more MCP-compatible tools (Amazon Q Developer and JetBrains AI Assistant have been mentioned in roadmap sessions), a visual MCP policy editor in the admin center, and possibly a Copilot Skill that lets users ask natural-language questions about Dataverse data directly in Teams.
The bigger story is that Dataverse is positioning itself as the data backbone for the multi-agent enterprise. By speaking MCP, it can serve data to whatever assistant a developer chooses—no longer locking them into a single vendor’s ecosystem. That openness, combined with hard governance wires, is what makes this announcement a turning point for AI-assisted development on the Microsoft stack.