Azure Pipelines MCP Server

A Model Context Protocol (MCP) server that provides AI assistants with deep knowledge and capabilities for Azure Pipelines.

This server enables LLMs (like Claude, Gemini, etc.) to:

• Validate pipeline YAML against the Azure DevOps API.
• Search for available tasks and their schema.
• Reference expressions, predefined variables, and YAML structure.
• Inspect Git repositories within your Azure DevOps project.

Features

• 🔍 Task Search & Reference: Look up task names, inputs, and documentation (e.g., "How do I use DotNetCoreCLI@2?").
• ✅ Pipeline Validation: Validate your YAML content using the official Azure DevOps Pipeline Preview API (detects syntax errors, invalid keywords, etc.).
• 📚 Documentation Access: Instant access to Azure Pipelines expressions, variables, and schema definitions.
• 📂 Repository Context: List available repositories to help context-aware pipeline creation.

Installation

Using npx (Recommended)

You can run the server directly using npx:

npx azure-pipelines-mcp

From Source

git clone https://github.com/sparx2206/azure-pipelines-mcp.git
cd azure-pipelines-mcp
npm install
npm run build
node dist/index.js

Configuration

To use the validation features and repository inspection, you must provide Azure DevOps credentials. These can be set as environment variables.

| Variable | Description | Required? | | :--- | :--- | :--- | | AZURE_DEVOPS_ORG | Your organization name (e.g., https://dev.azure.com/{org}). | ✅ Yes | | AZURE_DEVOPS_PAT | Personal Access Token with Build (Read & Execute) and Code (Read) scopes. | ✅ Yes | | AZURE_DEVOPS_PROJECT | Default project name. | ❌ Optional |

Claude Desktop

Add the following to your claude_desktop_config.json:

{
  "mcpServers": {
    "azure-pipelines": {
      "command": "npx",
      "args": ["-y", "azure-pipelines-mcp"],
      "env": {
        "AZURE_DEVOPS_ORG": "my-org",
        "AZURE_DEVOPS_PAT": "my-pat-token",
        "AZURE_DEVOPS_PROJECT": "my-project"
      }
    }
  }
}

VS Code (with MCP Extension)

If you are using an MCP extension in VS Code, configure it in .vscode/settings.json or the extension specific config:

{
  "mcp.servers": {
    "azure-pipelines": {
      "command": "npx",
      "args": ["-y", "azure-pipelines-mcp"],
      "env": {
        "AZURE_DEVOPS_ORG": "...",
        "AZURE_DEVOPS_PAT": "..."
      }
    }
  }
}

Available Tools

| Tool | Description | | :--- | :--- | | validate_pipeline_yaml | Validates YAML content using Azure DevOps API. | | search_pipeline_tasks | Searches for pipeline tasks (e.g., "docker"). | | get_task_reference | Gets detailed schema/inputs for a specific task. | | get_expressions_reference | Documentation for YAML expressions (functions, syntax). | | get_predefined_variables | List of system variables (Build.BuildId, etc.). | | get_yaml_schema | Reference for YAML structure (steps, stages, pool). | | get_repositories | Lists Git repositories in the project. | | get_dummy_pipeline | Utility to find a pipeline for validation context. | | create_dummy_pipeline | Utility to create a pipeline for validation context. |

Validation & Dummy Pipeline

The validate_pipeline_yaml tool uses the official Azure DevOps "Pipeline Preview" API. This API requires the context of an existing pipeline to resolve variable groups, environments, and repository resources correctly.

To facilitate this, the MCP server may need to create a placeholder pipeline in your project:

1. Name: DummyValidationPipeline
2. Folder: \AI\ (to keep it organized and separate from real pipelines)
3. Purpose: It acts as a target for the "Preview" API call. It is never executed by the MCP server.
4. Creation: The create_dummy_pipeline tool creates this pipeline pointing to azure-pipelines.yml in your default repository.

Note: If you delete this pipeline, the validation tool might prompt to recreate it.

Development

Setup

npm install

Testing

# Run unit tests
npm test

# Watch mode
npm run test:watch

Updating Data

The server uses embedded datasets for speed. To update them from Microsoft documentation:

npm run update-data

License

MIT