Azure DevOps MCP Server

An MCP (Model Context Protocol) server that exposes 41 Azure DevOps tools to AI assistants — covering pipelines, Git repositories, pull requests, releases, boards, work items, test management, and wikis.

Authentication: Azure Entra device code flow — no app registration required.

What It Can Do

Prerequisites

• Python 3.10+
• Access to an Azure DevOps organization (dev.azure.com)

Installation

From PyPI (recommended)

pip install azure-devops-mcp

Or with uv:

uv pip install azure-devops-mcp

From source

git clone <your-repo-url>
cd azure-devops-mcp
pip install -e .

Quick Start

After installing, follow these 3 steps:

Step 1: Install

pip install azure-devops-mcp

Step 2: Add to your AI editor

Open your editor's MCP config file and add the server. Here's VS Code as an example (.vscode/mcp.json):

Option A — With a PAT (simplest, no browser prompt):

{
  "servers": {
    "azure-devops": {
      "command": "azure-devops-mcp",
      "env": {
        "AZURE_DEVOPS_ORG_URL": "https://dev.azure.com/YOUR-ORG",
        "AZURE_DEVOPS_PAT": "YOUR-PERSONAL-ACCESS-TOKEN"
      }
    }
  }
}

To create a PAT: Azure DevOps → User Settings → Personal Access Tokens → New Token.

Option B — With Device Code Flow (no PAT needed):

First, log in once from your terminal:

export AZURE_DEVOPS_ORG_URL=https://dev.azure.com/YOUR-ORG
azure-devops-mcp auth login

This opens your browser, you sign in, and the token is cached locally. Then add to your editor — no PAT required:

{
  "servers": {
    "azure-devops": {
      "command": "azure-devops-mcp",
      "env": {
        "AZURE_DEVOPS_ORG_URL": "https://dev.azure.com/YOUR-ORG"
      }
    }
  }
}

Token is cached at ~/.azure-devops-mcp/token_cache.json and auto-refreshes (~90 days). Run azure-devops-mcp auth status to check your login state.

Step 3: Use it

Open your AI assistant (Copilot Chat, Claude, etc.) and ask:

List all pipelines in my project

The server handles authentication and API calls — the AI tool sees 41 Azure DevOps commands automatically.

Configuration

Set these environment variables (via your shell, .env file, or MCP client config):

Authentication

The server supports 4 authentication methods, auto-detected from environment variables:

Method 1: Personal Access Token (PAT) — simplest

Best for: scripts, personal use, CI/CD pipelines.

"env": {
  "AZURE_DEVOPS_ORG_URL": "https://dev.azure.com/my-org",
  "AZURE_DEVOPS_PAT": "your-pat-here"
}

Method 2: Service Principal (Client Credentials) — for apps

Best for: registered Azure AD applications, service-to-service, automated workflows.

"env": {
  "AZURE_DEVOPS_ORG_URL": "https://dev.azure.com/my-org",
  "AZURE_CLIENT_ID": "your-app-id",
  "AZURE_CLIENT_SECRET": "your-secret",
  "AZURE_TENANT_ID": "your-tenant-id"
}

Method 3: Managed Identity — for Azure-hosted environments

Best for: running on Azure VMs, App Service, Azure Functions, AKS. No credentials to manage.

pip install azure-devops-mcp[managed-identity]

"env": {
  "AZURE_DEVOPS_ORG_URL": "https://dev.azure.com/my-org",
  "AZURE_USE_MANAGED_IDENTITY": "true"
}

Method 4: Device Code Flow — interactive (default)

Best for: developers running locally. No PAT or app registration needed.

Recommended: Pre-authenticate from your terminal before configuring your AI editor:

export AZURE_DEVOPS_ORG_URL=https://dev.azure.com/my-org
azure-devops-mcp auth login    # Opens browser, caches token
azure-devops-mcp auth status   # Verify login state

The browser opens automatically. After you sign in, the token is cached at ~/.azure-devops-mcp/token_cache.json and auto-refreshes (~90 days).

If you skip auth login, the server will prompt via stderr on the first API call — but most AI editors don't display stderr, so pre-authenticating is strongly recommended.

Priority: PAT → Client Credentials → Managed Identity → Device Code. The first method detected wins.

Setting Environment Variables

There are 3 ways to pass your configuration:

Option A — In your AI editor's MCP config (recommended)

Every MCP-compatible editor has an "env" block in its config. This is the easiest approach — just add your variables there. See the Integration section below for editor-specific examples.

Option B — Via a .env file

Create a .env file in your working directory (use .env.example as a template):

AZURE_DEVOPS_ORG_URL=https://dev.azure.com/my-org
AZURE_DEVOPS_PAT=my-pat-token

The server loads it automatically on startup via python-dotenv.

Option C — Via shell environment (CI/CD, Docker, manual runs)

export AZURE_DEVOPS_ORG_URL=https://dev.azure.com/my-org
export AZURE_DEVOPS_PAT=my-pat-token
azure-devops-mcp

Integration

After installing via pip, the azure-devops-mcp command is available on your PATH. Configure your AI tool:

VS Code (GitHub Copilot / Copilot Chat)

Add to .vscode/mcp.json in your workspace (or User Settings JSON under mcp.servers):

{
  "servers": {
    "azure-devops": {
      "command": "azure-devops-mcp",
      "env": {
        "AZURE_DEVOPS_ORG_URL": "https://dev.azure.com/your-organization",
        "AZURE_DEVOPS_PROJECT": "your-default-project"
      }
    }
  }
}

Claude Desktop

Add to your Claude Desktop config file:

• Windows: %APPDATA%\Claude\claude_desktop_config.json
• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json

{
  "mcpServers": {
    "azure-devops": {
      "command": "azure-devops-mcp",
      "env": {
        "AZURE_DEVOPS_ORG_URL": "https://dev.azure.com/your-organization",
        "AZURE_DEVOPS_PROJECT": "your-default-project"
      }
    }
  }
}

Claude Code (Cortex Code)

claude mcp add azure-devops -- azure-devops-mcp

Then set environment variables in your shell or Claude config.

Cursor

Add to .cursor/mcp.json in your workspace:

{
  "mcpServers": {
    "azure-devops": {
      "command": "azure-devops-mcp",
      "env": {
        "AZURE_DEVOPS_ORG_URL": "https://dev.azure.com/your-organization",
        "AZURE_DEVOPS_PROJECT": "your-default-project",
        "AZURE_TENANT_ID": "your-tenant-id-or-organizations"
      }
    }
  }
}

Windsurf

Add to ~/.codeium/windsurf/mcp_config.json:

{
  "mcpServers": {
    "azure-devops": {
      "command": "azure-devops-mcp",
      "env": {
        "AZURE_DEVOPS_ORG_URL": "https://dev.azure.com/your-organization",
        "AZURE_DEVOPS_PROJECT": "your-default-project"
      }
    }
  }
}

Any MCP-compatible client

The server communicates over stdio. Point any MCP client at:

azure-devops-mcp

Running from source (alternative)

If you cloned the repo instead of installing via pip:

{
  "servers": {
    "azure-devops": {
      "command": "uv",
      "args": ["run", "python", "-m", "azure_devops_mcp"],
      "cwd": "/path/to/azure-devops-mcp",
      "env": {
        "AZURE_DEVOPS_ORG_URL": "https://dev.azure.com/your-organization",
        "AZURE_DEVOPS_PROJECT": "your-default-project"
      }
    }
  }
}

All 41 Tools

Pipelines & Build Artifacts

Releases & Approvals

Git Repositories

Pull Requests

Boards & Work Items

Test Management

Wikis

Example Prompts

List all pipelines in the "MyProject" project

Show me the last 5 runs for pipeline 42

List all active pull requests in the "backend-api" repository

Create a pull request from feature/login to main titled "Add login page"

Show me pending release approvals

Approve release approval 1234 with comment "Looks good, deploying to prod"

Query all active bugs assigned to me:
SELECT [System.Id], [System.Title] FROM workitems
WHERE [System.WorkItemType] = 'Bug' AND [System.State] = 'Active'
AND [System.AssignedTo] = @Me

Get the wiki page at /Runbooks/Deployment from the project wiki

Show test results for test run 567 — only failures

Compare the main and develop branches in the "frontend" repo

Development

Run the MCP dev inspector to test tools interactively:

uv run mcp dev src/azure_devops_mcp/server.py

Architecture

azure-devops-mcp/
├── src/azure_devops_mcp/
│   ├── __init__.py            # Package init + version
│   ├── __main__.py            # CLI entry point (azure-devops-mcp command)
│   ├── server.py              # MCP server — registers all 41 tools
│   ├── auth.py                # MSAL device-code auth (no app registration needed)
│   └── clients/
│       ├── pipelines.py       # Pipeline + build artifact API calls
│       ├── releases.py        # Classic release + approval API calls
│       ├── git.py             # Git repos, branches, commits, PRs
│       ├── boards.py          # Boards, work items, comments, saved queries
│       ├── tests.py           # Test runs, results, code coverage
│       └── wiki.py            # Wiki pages
├── .env.example               # Environment variable template
├── pyproject.toml             # Python project config (PyPI metadata + build)
└── README.md

All client modules include retry logic with exponential backoff for 429 Too Many Requests and 5xx server errors.

License

MIT