Exercise: Integrate MCP with Copilot

[github-actions]

Integrate MCP with Copilot

👋 Hey there @bernatcg! Welcome to your Skills exercise!

This exercise will help you learn how to expand GitHub Copilot's capabilities with Model Context Protocol (MCP).

---

✨ This is an interactive, hands-on GitHub Skills exercise!

As you complete each step, I’ll leave updates in the comments:

• ✅ Check your work and guide you forward
• 💡 Share helpful tips and resources
• 🚀 Celebrate your progress and completion

Let’s get started - good luck and have fun!

_{— Mona}

_{If you encounter any issues along the way please report them here.}

[github-actions]

Step 1: Introduction to MCP and environment setup

In the Getting Started with GitHub Copilot exercise, we were introduced to the Mergington High School's extracurricular activities website, which allowed students to sign up for events.

And now we have a problem... but.. it's a good one! More teachers are asking to use it! 🎉

Our fellow teachers have lots of ideas but we can't seem to keep up with all the requests! 😮 To fix this issue, lets give GitHub Copilot an upgrade by enabling Model Context Protocol (MCP). To be more specific, we will add the GitHub MCP server, which will enable a combined workflow of issue management and website upgrades. 🧑‍🚀

Let's get started!

📖 Theory: What is Model Context Protocol (MCP)?

Model Context Protocol (MCP) is often referred to as "USB-C for AI" - a universal connector that allows GitHub Copilot (and other AI tools) to seamlessly interact with other services.

Essentially, it is a way to describe the capabilities and requirements of a service, so AI tools can easily determine what methods to use and to accurately provide the parameters. An MCP server is providing that interface.

graph LR
    A[Developer] -->|Uses| B[GitHub Copilot]
    B -->|Unified API| MCP[Model Context Protocol]

    MCP <-->|Unique API| C[(GitHub)]
    MCP <-->|Unique API| D[(Slack)]
    MCP <-->|Unique API| E[(Figma)]

    style B fill:#4CAF50,stroke:#333,stroke-width:2px

    subgraph "Less Context Switching, More Coding"
        B
        MCP
        C
        D
        E

    end

Loading

⌨️ Activity: Get to know your environment

Before we dive into MCP, let's start up our development environment and refamiliarize ourself with the extracurricular activity application.

1. Right-click the below button to open the Create Codespace page in a new tab. Use the default configuration.

   

2. Validate the Copilot Chat and Python extensions are installed and enabled.

   
   
   

3. Verify our application runs before modification. In the left sidebar, select the Run and Debug tab and then press the Start Debugging icon.

   📸 Show screenshot

   
   
   🤷 Having trouble?

   
   

   If the Run and Debug area is empty, try reloading VS Code: Open the command palette (Ctrl+Shift+P) and search for Developer: Reload Window.

   

4. Use the Ports tab to find the webpage address, open it, and verify it is running.

   📸 Show screenshot

   
   

   

⌨️ Activity: Add the GitHub MCP server

1. Inside your codespace, open the Copilot Chat panel and verify Agent mode is selected.

   
   Agent mode missing?

   
   

   • Verify VS Code is at least v1.99.0.

   • Verify the Copilot extension is at least v1.296.0.

   • Check if Agent mode is enabled in your user or workspace settings.

     

2. Inside your codespace, navigate to the .vscode folder, and create a new file named mcp.json. Paste the following contents:

   📄 .vscode/mcp.json

   {
     "servers": {
       "github": {
         "type": "http",
         "url": "https://api.githubcopilot.com/mcp/"
       }
     }
   }

3. In the .vscode/mcp.json file, click the Start button and accept the prompt to authenticate with GitHub. This has just informed GitHub Copilot of the MCP server's capabilities.

   

   
   

   

4. In the Copilot side panel, click the 🛠️ icon to show the additional capabilities.

   

5. Commit and push the .vscode/mcp.json file to the main branch.

   🪧 Note: Pushing directly to main is not a recommended practice. It is only to simplify this exercise.

6. Now that your MCP server configuration is pushed to GitHub, Mona should already be busy checking your work. Give her a moment and keep watch in the comments. You will see her respond with progress info and the next lesson.

Note

The next steps will involve creating GitHub issues. If you would like to avoid notification emails, you can unwatch the repository.

Having trouble?

Make sure:

• Your .vscode/mcp.json file is similar to the example provided.
• You pushed the changes to the main branch.

[github-actions]

Please, follow the steps above.
I'll watch your progress in the background to provide feedback. 🧐

[github-actions]

🎉🎉🎉 Nice work! Everything is perfect! 🎉🎉🎉
Preparing content for step 2! One moment... 🤓

[github-actions]

Step 2: Agent Mode and an MCP Server for GitHub

Great work! You just connected your first MCP server to GitHub Copilot! 🎉

🚨 The teachers have been busy opening new issues in your repository with bugs and feature requests! Go take a look - so many good ideas!

We should probably look into them and start researching for other upgrades. Fortunately, with an MCP server for GitHub, triaging these and even doing some research to get ahead should be pretty quick! 🕵️

📖 Theory: How MCP Tool Calling Works in Agent Mode

Now that we have the GitHub MCP connected, let's look at how agent mode actually uses these external tools.

With every prompt you send, Copilot also includes the catalog (list + schema) of available tools. Copilot can then plan and decide:

• Is any tool needed for this request?
• Which tool(s) best match the intent?
• What arguments (per each tool's input schema) should be passed?

Copilot then executes the chosen tool call(s) and streams results back to the LLM.

Tip

You can also explicitly nudge Copilot to call a specific tool by including #<tool_name> in your prompt (e.g #create_pull_request, #codebase).

From here, Copilot can use a set of GitHub‑aware tools to do more than just read or edit code in your repo. Here are a few things you can ask it to do:

• Discover similar public projects to get inspiration.
• Search issues considering description, comments, and likes.
• Turn the new ideas you like into issues right away so you don’t lose them.
• Retrieve an issue, make changes on a branch, and start a pull request.

Isn't that cool?! Now let's do it! 👩‍🚀

⌨️ Activity: Quickly find and save ideas

Let's put the GitHub MCP server to use by researching, comparing, and capturing enhancement ideas!

1. Close any open files inside your codespace. This will help reduce unnecessary context.

2. Ensure the Copilot Chat panel is open and Agent mode is selected. Verify the MCP server tools are also still available.

3. Ask Copilot to search GitHub for projects similar to this one.

   

   Search for any other repositories for organizing extra curricular activities
   

4. When an MCP tool is required, Copilot may ask for permission. Verify the request and modify if necessary, then click Continue.

   

5. Ask Copilot to describe one of the projects. Explore until you find something you like.

   

   Please look at the code for the 3rd option and give me a detailed description of the features.
   

6. Use Copilot to compare and generate ideas for enhancements.

   

   Please compare these features to our project. Which would be new?
   

7. Nice! Let's have Copilot create issues to save these ideas.

   

   I like it. Let's create issues for these in my repository.
   

8. Copilot will ask for permission to create issues on your repository. Click Continue for each new issue. Reminder: verify the request before running.

   

9. Since we are done researching, let's finish this chat session to clear the context. At the top of the Copilot Chat panel, click the New Chat icon (plus sign).

10. With the new issues created, Mona should already be busy checking your work. Give her a moment and keep watch in the comments. You will see her respond with progress info and the next lesson.

Note

The Model Context Protocol (MCP) landscape is quickly evolving. Many servers, including the Official GitHub MCP server are in active development and do not have full parity with their stable APIs.

[github-actions]

Please, follow the steps above.
I'll watch your progress in the background to provide feedback. 🧐