MCP Dataverse

What codeurali/mcp-dataverse MCP server does

The codeurali/mcp-dataverse MCP server connects MCP-compatible clients to Microsoft Dataverse. It gives agents access to live environment data instead of requiring them to infer table names, columns, relationships, or query syntax. The repository documents 79 tools, four resources, and 10 guided workflows.

Supported operations cover table and schema metadata, option sets, relationships, entity keys, OData queries, FetchXML, pagination, and full record lifecycle actions. Agents can also associate records, execute bound and unbound Dataverse actions and functions, run atomic batches of up to 1,000 operations, publish customizations, create sitemaps, and use relevance search.

Additional areas include users, teams, roles, role privileges, files and images, annotations, audit logs, plug-in and workflow traces, custom actions, plug-ins, environment variables, connection references, columns, custom tables, relationships, record sharing, record merging, views, business units, impersonation, and duplicate detection.

How it works

The codeurali/mcp-dataverse MCP server uses the Model Context Protocol to present Dataverse functionality as individual tools. Responses use a structured shape containing a summary, data, and suggestions. Destructive operations require explicit confirmation, and cached credentials are protected with AES-256-GCM rather than written to logs.

For local development, device-code authentication starts when the first tool is called. The user retrieves the sign-in code from the MCP output, submits it at Microsoft's device-login page, and then continues with an encrypted cached token. Client credentials are intended for CI/CD, Docker, and other unattended workloads. Managed identity is intended for Azure-hosted services such as App Service and Container Apps.

The server can use standard local MCP communication or HTTP transport. HTTP mode exposes an /mcp endpoint and can protect requests with a bearer secret.

Setup and configuration

Node.js 20 or newer is required. The repository provides an interactive setup command:

npx mcp-dataverse install

The wizard configures the environment, registers the server in VS Code, and starts Microsoft account authentication. Client-specific configuration is required for clients such as Claude, Cursor, and Windsurf. The repository also provides an update command and a diagnostic command for setup or connectivity issues.

HTTP operation is configured with MCP_TRANSPORT=http, MCP_HTTP_PORT, and MCP_HTTP_SECRET, followed by starting the built server. Client credentials and managed identity require the corresponding authentication configuration described by the project documentation.

Tools and capabilities

The documented capability groups include:

• Metadata inspection and schema operations
• OData, FetchXML, paginated retrieval, and relevance search
• Create, read, update, delete, upsert, assign, and merge operations
• Relationship association, disassociation, and association queries
• Actions, functions, workflows, and atomic batches
• Users, teams, roles, privileges, sharing, and impersonation
• File and image columns, notes, attachments, and trace data
• Solutions, customizations, columns, tables, relationships, and environment settings

Tool selection can be reduced in the MCP client when an agent only needs a subset of these capabilities.

Limitations and notes

The server is comprehensive, so exposing every tool may make agent tool selection less focused. The README recommends deselecting tools that are not needed. Device-code authentication requires interactive sign-in and access to the MCP output channel; client credentials or managed identity are better suited to unattended deployments. The HTTP example uses a local secret and should be configured according to the deployment's security requirements.