Supercharge GitHub Copilot's Agent Mode

21 May 2025
MCP,
Agent,
VSCode,
Secret,
Token

Visual Studio Code docs on MCP servers and Agents #

The Visual Studio Code docs on MCP servers show how to supercharge GitHub Copilot's agent mode by plugging in external tools using the Model Context Protocol (MCP).

What is MCP? #

MCP is an open standard that lets AI communicate with tools and services via a unified interface. With MCP-compatible servers hooked up to VS Code, Copilot can do more than just suggest code – it can:

Run file operations
Access databases
Make API calls

All directly inside your editor.

The Guide Covers: #

Setting up MCP servers
Configuring them in VS Code
Managing the whole integration

It also explains what types of servers are supported and how to use MCP tools in agent mode.

The goal?

Smoother workflows, smarter tooling, and more time for building cool stuff.

👉 Dive into the docs: VS Code MCP Guide 📚

Sample MCP Servers I Use #

Here are a few MCP servers I currently have configured:

Atlassian integration (Jira & Confluence): sooperset/mcp-atlassian
NX integration for monorepos: nx-mcp

More MCP servers are available here: 👉 List of MCP Servers

Secure Setup Tips #

When setting up your mcp.json, you might worry about exposing settings. To hide your MCP server variables securely, follow this flow:

{
  "inputs": [{
    "id": "atl_token",
    "type": "promptString",
    "password": true,
    "description": "Enter your Atlassian Jira and Confluence token"
  },{
    "id": "BS_user",
    "type": "promptString",
    "password": false,
    "description": "Enter your BS User"
  },{
    "id": "BS_token",
    "type": "promptString",
    "password": true,
    "description": "Enter your BS token"
  }],
}

This will ask for secrets at server start and remember them. Use placeholder variables in your config (e.g., ${input:atl_token}).

"CONFLUENCE_API_TOKEN": "${input:atl_token}",

Project-Specific Instructions for Copilot #

GitHub Copilot now supports a .github/copilot-instructions.md file in your repo to tailor responses to your tools, workflows, and coding style.

How to Set It Up #

Create .github/copilot-instructions.md in your repository root
Write clear, natural-language instructions in Markdown

Copilot will now include these automatically in its responses.

How to Check It’s Working #

Look for .github/copilot-instructions.md in the References section of Copilot Chat responses.

👉 GitHub Docs – Copilot Instructions

Custom Prompts (Optional) #

You can also create prompt files like .github/prompts/prompts_Example.md. These files guide Copilot to align with your coding style, frameworks, and project needs, enhancing its suggestions. For more details, visit the official documentation.

Additional specialized Instructions #

Additionally two files are available that you can configure in your Settings.json with the following settings. The names alone explain what they are for.

"github.copilot.chat.codeGeneration.instructions": [
  {
    "file": ".github/code-style.md"
  }
],
"github.copilot.chat.testGeneration.instructions": [
  {
    "file": ".github/test-style.md"
  }
]

Previous: Replacing MUI - A New Design System Approach