AI & Automation••3 phút đọc

Automate Workflow: Cursor + Azure DevOps with MCP Server

Modern software development often feels like a constant battle against context switching. A developer starts their day by reading a ticket on Azure DevOps (ADO), manually creating a local branch, setting up boilerplate code, and eventually pushing a Pull Request (PR). This manual "plumbing" consumes 30-40% of a developer's cognitive energy before they even write a single line of business logic.

What if you could simply mention a ticket ID in your IDE and let an AI agent handle the rest? By combining Cursor (the AI-native IDE) with the Model Context Protocol (MCP) and Azure DevOps, we can build a seamless "Ticket-to-PR" pipeline.

The Architecture: Cursor, MCP, and Azure DevOps

The solution relies on a three-tier architecture that bridges the gap between your local coding environment and your project management tools:

  • Cursor (The Brain): Acts as the primary interface. It understands the context of your codebase and executes instructions using Large Language Models (LLMs).
  • MCP Server (The Bridge): An implementation of Anthropic’s Model Context Protocol. It exposes specific "Tools" (like fetching work items or creating PRs) that Cursor can invoke.
  • Azure DevOps (The Source of Truth): Provides the REST APIs for Work Items, Repositories, and Pull Requests.

The Automated Workflow

With this setup, the developer experience transforms into a high-level orchestration:

  1. Mention: In Cursor, you type @ADO-123 implement this feature.
  2. Fetch: The MCP Server calls the ADO REST API to retrieve the title, description, and acceptance criteria for ticket 123.
  3. Execute: Cursor analyzes the requirements, creates a new feature branch (feature/ADO-123-title), generates the code, and writes unit tests.
  4. Finalize: Cursor uses an MCP tool to push the branch and open a Pull Request, linking it back to the original work item.

Technical Implementation: Building the MCP Server

The MCP server is typically a Node.js or Python application. Below is a conceptual example of how to implement a tool that fetches work item details using the Azure DevOps Node API.


import { McpServer } from "@modelcontextprotocol/sdk/server/mcp.js";
import * as azdev from "azure-devops-node-api";

const server = new McpServer({
  name: "AzureDevOps-Bridge",
  version: "1.0.0"
});

// Setup Azure DevOps Connection
const orgUrl = "https://dev.azure.com/YourOrg";
const token = process.env.ADO_PAT;
const authHandler = azdev.getPersonalAccessTokenHandler(token);
const connection = new azdev.WebApi(orgUrl, authHandler);

// Register the "get_work_item" tool
server.tool("get_work_item", { id: "number" }, async ({ id }) => {
  const witApi = await connection.getWorkItemTrackingApi();
  const workItem = await witApi.getWorkItem(id);
  
  return {
    content: [{
      type: "text",
      text: JSON.stringify({
        title: workItem.fields?.["System.Title"],
        description: workItem.fields?.["System.Description"],
        criteria: workItem.fields?.["Microsoft.VSTS.Common.AcceptanceCriteria"]
      })
    }]
  };
});

Strategic Insights & Challenges

While the "AI Dev Agent" model provides a massive productivity boost, it is not without hurdles:

1. The Quality of Input

AI is only as good as the instructions it receives. If a ticket has a vague description, the AI will likely implement the wrong logic. Organizations must enforce strict Acceptance Criteria standards to make this automation viable.

2. Security and Guardrails

Giving an AI agent write access to your repository requires caution. It is recommended to use Personal Access Tokens (PAT) with limited scopes and always require a human to review the final PR before merging.

3. Context Window Management

For large legacy codebases, the AI might struggle to understand the impact of changes. Combining this workflow with a RAG (Retrieval-Augmented Generation) system for internal documentation can help the agent navigate complex architectures.

The Future of AI-Native Development

The goal isn't to replace developers but to eliminate the "toil" of software engineering. By building an internal AI Dev Agent, teams can reduce the time-to-market for features and ensure that naming conventions, testing standards, and workflow steps remain consistent across the entire organization.

← Quay lại Blog