Getting Started

Quick start guide for MCP Dataverse — an MCP server exposing 79 AI-callable tools for Microsoft Dataverse.

Prerequisites

  • Node.js 20+
  • A Dataverse environment with a licensed user account (e.g. https://yourorg.crm.dynamics.com)
  • An MCP-compatible client (VS Code + Copilot, Claude Desktop, Cursor, etc.)

Installation

Recommended — Interactive CLI 

npx mcp-dataverse install

The wizard will:

  1. Ask for your Dataverse environment URL
  2. Save configuration to ~/.mcp-dataverse/config.json
  3. Register the server in VS Code via code --add-mcp
  4. Authenticate with your Microsoft account (device code flow)

Manual Configuration

Create ~/.mcp-dataverse/config.json:

{
  "environmentUrl": "https://yourorg.crm.dynamics.com",
  "requestTimeoutMs": 30000,
  "maxRetries": 3
}

See multi-client-setup.md for client-specific configuration.

Configuration Options

Option Type Default Description
environmentUrl string — Your Dataverse org URL (must be https://*.dynamics.com)
requestTimeoutMs number 30000 HTTP request timeout in milliseconds
maxRetries number 3 Retry count on transient errors (0–10)

Environment Variables

All config values can be set via environment variables (useful for MCP client env blocks):

Variable Maps to
DATAVERSE_ENV_URL environmentUrl
REQUEST_TIMEOUT_MS requestTimeoutMs
MAX_RETRIES maxRetries
MCP_CONFIG_PATH Path to a custom config.json file

Priority: environment variables > MCP_CONFIG_PATH > ~/.mcp-dataverse/config.json > ./config.json.

Authentication

Three methods are supported — choose based on where the server runs:

→ Authentication overview & setup guide

Quick start (device code, local use): authentication triggers on the first tool call.

  1. Check VS Code Output panel → MCP for a device code and sign-in URL
  2. Open https://microsoft.com/devicelogin, paste the code, and log in with your work account

Tokens are cached encrypted (AES-256-GCM) and renewed silently (~90 days).

Running the Server

Stdio transport (default)

Used by VS Code, Claude Desktop, Cursor, and most MCP clients:

npx mcp-dataverse

HTTP transport

For clients that connect over HTTP (or for shared/remote access):

npx mcp-dataverse --transport http --port 3001

The MCP endpoint is available at http://localhost:3001/mcp.

HTTP Transport

The server can run in HTTP mode instead of stdio, which enables multi-client connections and browser-based access.

Starting the HTTP server

Set the MCP_TRANSPORT environment variable to http:

MCP_TRANSPORT=http MCP_HTTP_PORT=3000 node dist/server.js

Or in config.json:

{
  "transport": "http",
  "httpPort": 3000
}

Environment variables

Variable Default Description
MCP_HTTP_PORT 3000 Port to listen on
MCP_HTTP_JSON_RESPONSE true Enable JSON response mode (vs SSE streaming)
MCP_HTTP_SECRET (none) Bearer token for authentication. If set, all /mcp requests must include Authorization: Bearer <token>
BEARER_TOKEN (none) Deprecated alias for MCP_HTTP_SECRET. Use MCP_HTTP_SECRET in new setups.
MCP_HTTP_CORS_ORIGIN * Allowed CORS origin. Set to your app origin (e.g. http://localhost:5173) when auth is enabled to restrict browser access.

Authentication

When MCP_HTTP_SECRET is set, every request to /mcp must include:

Authorization: Bearer your-secret-token

Health check 

GET /health

Returns {"status":"ok","version":"...","tools":<count>}. No authentication required.

Multi-client support

Each client gets a dedicated MCP session identified by mcp-session-id. Sessions are automatically cleaned up on disconnect.

Diagnostics

Run the built-in health check to verify your configuration and connectivity:

npx mcp-dataverse doctor

This checks:

  • Configuration file validity
  • Dataverse environment reachability
  • Authentication token status
  • API connectivity (WhoAmI)

Performance Tip

MCP Dataverse ships 60+ tools to cover the full Dataverse API surface. Most AI models perform best with fewer tools loaded in context.

Recommendation: In VS Code, open the Chat panel and deselect the tools you don’t need for your current task. This reduces noise and helps the agent pick the right tool faster.

You can also call dataverse_suggest_tools with a short description of what you want to do — it returns only the relevant tools.

Next Steps