AI 코딩 에이전트 (Claude Code, GitHub Copilot, Codex CLI, Cursor, Gemini CLI)를 제공하는 최소 MCP 서버는 Dataverse Web API v9.2을 통해 Microsoft Dataverse / XRM 환경에 대한 액세스를 읽습니다.

이 존재 이유: Microsoft는 자체 Dataverse MCP 서버를 발송하지만, Power Platform 관리자 설정, 관리 환경 및 통화 당 Copilot 크레딧이 필요합니다. xrm-mcp]는 기존 Azure CLI 또는 Microsoft 계정을 사용하여 모든 Dataverse 환경에서 작동하는 드롭 인 대안입니다. 더 읽기 →

특징

• ** 마이크로 소프트 MCP 청구 ** - 직접 Dataverse Web API를 호출, 코파이트 크레딧이 소모되지 않습니다 -No admin toggles — 로그인할 수 있는 모든 Dataverse 환경과 함께 작동합니다.
• 필요한 관리 환경 — 표준 환경 일 벌금 -No per-environment setup — org url은 모든 도구 호출의 매개 변수입니다.
• Multi-tenant by design - 같은 세션에서 여러 개의 orgs에 연결하여 per-environment identity 캐싱을 사용하여 10ants 사이 전환이 필요하지 않습니다.
• **Azure CLI + MSAL auth ** - 트리스 az first, 상호 작용 장치 흐름으로 돌아갑니다

xrm-mcp

A minimal MCP server that gives AI coding agents (Claude Code, GitHub Copilot, Codex CLI, Cursor, Gemini CLI) clean read + careful write access to Microsoft Dataverse / XRM environments via the Dataverse Web API v9.2.

Why this exists: Microsoft ships its own Dataverse MCP server, but it requires Power Platform admin setup, Managed Environments, and Copilot Credits per call. xrm-mcp is a drop-in alternative that works with any Dataverse environment using your existing Azure CLI or Microsoft account. Read more →

Features

• No Microsoft MCP billing — calls the Dataverse Web API directly, no Copilot Credits consumed
• No admin toggles — works with any Dataverse environment you can log into
• No Managed Environment required — standard environments work fine
• No per-environment setup — org_url is a parameter on every tool call
• Multi-tenant by design — connect to multiple orgs in the same session, with per-environment identity caching so switching between tenants doesn’t require re-authenticating each time
• Azure CLI + MSAL auth — tries az first, falls back to interactive device flow
• 8 MCP tools — ping, find/list tables, describe schema, query, create, update, upsert

Installation

For uv, pipx, virtual environments, pinned tags, and client setup, see the installation guide.

Install via pipx (recommended):

pipx install git+https://github.com/jukkan/xrm-mcp.git

Or via pip:

pip install git+https://github.com/jukkan/xrm-mcp.git

Usage

Configuring your AI agent

Add xrm-mcp to your agent’s MCP configuration.

Claude Desktop / Claude Code (~/.claude/claude_desktop_config.json or .mcp.json):

{
  "mcpServers": {
    "xrm-mcp": {
      "command": "xrm-mcp"
    }
  }
}

GitHub Copilot (VS Code) — add to .vscode/mcp.json in your workspace:

{
  "servers": {
    "xrm-mcp": {
      "type": "stdio",
      "command": "xrm-mcp"
    }
  }
}

Once configured, pass your environment URL with every request and the agent takes it from there:

“Show me all project records from https://myorg.crm4.dynamics.com created in the last 30 days”

Running the server manually

xrm-mcp

The server uses FastMCP and communicates over stdin/stdout (stdio transport).

Authentication

XRM MCP attempts authentication in the following order:

1. Azure CLI — if az is available and logged in
2. MSAL device flow — interactive browser-based login

Tokens are cached at ~/.xrm-mcp/cache.json.

If you work across multiple tenants (e.g. your own production environment, demo environments, and customer tenants), XRM MCP remembers which identity last worked for each org_url in ~/.xrm-mcp/identity_cache.json and prefers it on