5 Steps to Set Up Atlassian MCP for VS Code

VS Code reads MCP server configuration from your settings. Open your VS Code settings JSON and add the Atlassian MCP server entry pointing at https://mcp.atlassian.com/v1/mcp. The first time you run a prompt that triggers the MCP connection, VS Code opens a browser window for OAuth 2.1 consent — you authenticate with your Atlassian account, grant access to your Jira site, and you're connected. No API tokens to manage, no credentials in your repo.

With the MCP connection live, you can create Jira issues without leaving your editor. Start with a project plan file — a simple markdown document listing your epic, stories with acceptance criteria, and test tickets. Then ask Copilot to read the plan and create everything in Jira. For this demo, I wrote a plan for a "Next.js Product Landing Page" with six stories — scaffold, hero section, features grid, testimonials, footer with CTA, and responsive dark mode — plus three test tickets for visual regression, responsive breakpoints, and accessibility. One prompt, one plan file, and the entire Jira board is populated with acceptance criteria written at the quality you'd hand to a contractor — exactly what Step 3 from Part 1 demands.

In Part 1, Step 1 was "onboard the agent like you would a new hire." VS Code with GitHub Copilot reads an AGENTS.md file in your project root for exactly this purpose. Create one that tells Copilot your project's Jira key, your tech stack conventions, and your definition of done. Point it at the Atlassian MCP server so it knows to pull requirements from Jira rather than inventing them. I've included a full template below — the key sections are project overview, Jira workflow rules, and coding conventions. Once this file exists, every prompt Copilot receives carries your team's standards.

This is where Part 1's Step 4 comes alive — "let the agent execute against requirements, not conversation." Instead of describing what you want in a prompt, you point Copilot at a ticket. Ask it to read a specific Jira story via MCP, understand the acceptance criteria, and implement accordingly. Copilot pulls the story details — title, description, acceptance criteria — and develops against them. If the story says the hero section needs a headline, subheadline, CTA button, and responsive layout, that's exactly what gets built. The requirements live in Jira, not in your conversation history.

Part 1's final step was "close the loop with the same rigour you'd expect from any developer." Here's what that looks like in practice. Ask Copilot to check the implementation against the story's acceptance criteria, run your tests, and move the ticket to Done in Jira — all via the same MCP connection. No ticket closes without the definition of done being met. If something's missing, the ticket stays open and you iterate. The entire lifecycle — plan, build, validate, close — happens from your editor through a single MCP connection.

The setup takes about fifteen minutes. After that, you have a workflow where your AI agent operates with the same discipline as any developer on your team — tracked work, clear requirements, and a definition of done enforced at every step. Next time, I'll walk through the same setup with Claude Code, which takes a fundamentally different approach to the same problem.