Github Copilot - MCP
- Radek Stolarczyk
- 2 days ago
- 6 min read
Your AI Workflow with GitHub MCP Server
Overview
GitHub MCP Server helps you connect AI (like Copilot) to your tools and workflows without complicated setup.
Instead of configuring everything manually (Docker, tokens, configs), it gives you a ready-to-use bridge between AI and your development environment.
Think of it as making AI actually usable everywhere—not just in your IDE.
What is MCP?
MCP (Model Context Protocol) is a standard that lets AI tools connect to external systems like:
Code repositories
APIs
Databases
Developer tools
A good way to think about it:
MCP = “USB-C for AI tools”
Just like USB-C works across many devices, MCP works across:
Different AI tools
Different environments
Different data sources
This means you don’t have to rebuild integrations every time.
How MCP Works (Simple Explanation)
There are three main ways MCP connects things:
1. Local → Local
Your AI talks to a server on your machine, which accesses local files or data.
✔ Best for:
Private work
Fast access
2. Local → Remote (Bridge)
Your AI still uses a local server, but that server talks to online services (APIs).
✔ Best for:
Secure API calls
Controlled access to external systems
3. Remote → Remote
Everything happens in the cloud.
✔ Best for:
SaaS tools
Large-scale systems
Why GitHub MCP Server Matters
Without MCP, setting up AI workflows can be annoying:
You need Docker
You manage tokens manually
You configure everything yourself
GitHub MCP Server removes all of that.
You get:
One-click authentication
No manual setup
Works on web, mobile, and desktop
It’s basically plug-and-play AI for GitHub.
What You Can Actually Do With It
Once connected, MCP lets Copilot go beyond just suggesting code.
You can:
Analyze your entire codebase
Summarize projects
Create and manage issues
Open and update pull requests
Automate repetitive tasks
It even supports semantic search, meaning It understands your code, not just text
Real Example
Instead of manually doing this:
Read code
Find issues
Create tasks
Prioritize work
You can just ask:
“Analyze this repo and create issues for improvements”
And MCP + Copilot can handle it.
Why This is Powerful
The biggest advantage is consistency + scalability:
Works across environments (web, mobile, desktop)
Scales automatically
Integrates into existing workflows
So instead of AI being a “tool you occasionally use,” it becomes:
Part of your entire development workflow
Final Takeaway
GitHub MCP Server is important because it:
Removes setup friction
Connects AI to real tools
Enables automation at scale
Simple Mental Model
MCP Server =“The connector that lets AI actually do real work in your projects”
Configure, Connect, and Use GitHub MCP Server in VS Code
Overview
This unit is all about getting MCP working inside VS Code so you can use AI (Copilot) directly in your coding environment.
Goal:Bring AI-powered workflows into your editor, not just the browser.
The Big Picture
When you set up MCP in VS Code, you’re essentially:
Connecting VS Code → MCP Server
Authenticating with GitHub
Enabling Copilot to:
Access your repo
Automate tasks
Assist across workflows
Option 1: Setup Using OAuth (Recommended)
This is the simplest and most common method.
Steps (What’s actually happening)
Open Command Palette
Ctrl+Shift+P / Cmd+Shift+P
Run:
MCP: add server
Choose:
HTTP (or Server-Sent Events)
Enter server URL:
Choose where to save config:
User settings → works everywhere
Workspace → project-specific
Sign in with GitHub (OAuth)
What OAuth Does
OAuth = secure login via GitHub
No manual tokens
Automatic authentication
Easiest setup
Result
After setup:
VS Code is now connected to MCP Copilot can:
Analyze code
Manage issues
Automate workflows
Option 2: Setup Using Personal Access Token (PAT)
Use this when you need more control or customization.
When to Use PAT
Enterprise environments
Custom permissions
Restricted OAuth usage
Key Idea
Instead of logging in automatically, you:
Provide a token manually
What You Add (Conceptually)
You configure:
Authorization header
Secure input prompt
This tells VS Code:
“Use this token to authenticate instead of OAuth”
Important
Token must have correct permissions (like repo)
Stored securely via prompt (not hardcoded)
Option 3: Local MCP Server (Docker – Advanced)
This is optional and more advanced.
When You Need This
Enterprise restrictions
Custom environments
Full control over server
What Happens Here
Instead of using GitHub’s hosted MCP:
You run MCP locally using Docker
Requirements
Docker installed and running
PAT (OAuth doesn’t work here)
What It Enables
More control
Custom environments
Enterprise compatibility
Using MCP in VS Code
Once everything is set up:
MCP works behind the scenes with Copilot
You can:
Ask Copilot to analyze your repo
Generate code or tests
Create issues or PRs
Automate workflows
All without leaving VS Code
Troubleshooting (Very Important for Exams)
If something doesn’t work, check these first:
Common Issues
1. Not signed in
Fix: Re-authenticate GitHub in VS Code
2. Token problems (PAT)
Fix:
Check scopes
Re-enter token
3. Configuration errors
Fix:
Look for typos
Verify JSON config
4. Docker issues (local setup)
Fix:
Ensure Docker is running
5. Temporary glitches
Fix:
Restart VS Code
Restart MCP server
Simple Mental Model
Setting up MCP in VS Code =“Connecting your editor to an AI-powered backend that can act on your code”
Exam Tips (Important)
They’ll likely test:
Difference between OAuth vs PAT
When to use Docker setup
What MCP enables inside VS Code
Basic setup flow (Command Palette → MCP server → URL)
One-Line Summary
MCP in VS Code lets Copilot do real work in your project, not just suggest code
GitHub MCP Server with Copilot Chat
Overview
This unit is where everything comes together:
Copilot + Agent Mode + MCP = real automation
Instead of just answering questions, Copilot can now:
Plan tasks
Execute them
Iterate and improve
What Changes with Agent Mode?
Normally, Copilot works like this:
You ask → It answers
But in Agent Mode, it becomes:
You ask → It plans → It acts → It improves
What “Agentic” Actually Means
Agentic capabilities mean Copilot can:
Work through multi-step tasks
Decide which tools to use
Adapt based on results
Improve its output over time
It behaves more like a teammate, not just a tool.
How MCP Fits Into This
On its own, Agent Mode is powerful.
But it’s limited to what’s inside your editor.
When MCP is added:
Copilot can now:
Access external tools (GitHub, APIs, etc.)
Pull in more context (beyond your code)
Perform actions across systems
MCP basically gives Agent Mode access to the outside world
What This Looks Like in Practice
Without MCP:
“Explain this code”
With MCP + Agent Mode:
“Analyze this repo, create issues, and suggest improvements”
And Copilot can:
Read the repo
Analyze structure
Create issues
Suggest fixes
How to Use It in VS Code
Steps
Open Copilot Chat
Switch to Agent Mode
Click Select tools
Choose MCP tools
Start prompting
Example Prompts
“Summarize this repository”
“Create issues for performance improvements”
“Analyze this codebase and suggest refactoring”
The Big Upgrade (Important Concept)
MCP + Agent Mode enables something called:
Agentic loops
What is an Agentic Loop?
Instead of one response, Copilot:
Gathers information
Analyzes it
Takes action
Reviews results
Improves output
It keeps going until the task is done properly
Key Benefits
1. More Context
Copilot can use:
Your code
External tools
Repository data
Better understanding → better results
2. Less Manual Work
Copilot can:
Create issues
Manage workflows
Run tasks
You focus on decisions, not repetition
3. Seamless Workflow
No need to:
Switch apps
Manually connect tools
Everything happens in one flow
Best Practices (Very Important)
1. Be Clear About Goals
Instead of:
“Help with this repo”
Say:
“Analyze this repo and create issues for bugs and missing tests”
2. Provide Context
Mention files
Add links
Explain background
More context = better results
3. Set Boundaries
You can control behavior:
“Only plan, don’t make changes yet”
“Do not modify files”
4. Ask for a Plan First
Before big changes:
“Show me your plan before executing”
5. Use Instructions / Prompt Files
Define rules once
Keep behavior consistent
Common Mistake
Thinking Copilot will “just figure it out”
It works best when:
You guide it
You refine it
You iterate
Simple Mental Model
MCP + Agent Mode =“Copilot that can think, act, and use tools like a developer”
One-Line Summary
This setup turns Copilot from a helper into a workflow executor
Exam Tip
They may test:
What “agentic” means
Why MCP improves Agent Mode
What “agentic loop” is
Difference between normal Copilot vs Agent Mode