MCP Server: Awesome Copilot

This is an MCP server that retrieves GitHub Copilot customizations from the awesome-copilot repository.

Install

Prerequisites

.NET 10 SDK
Aspire CLI nightly: iex "& { $(irm https://aspire.dev/install.ps1) } -Quality dev"
Visual Studio Code with
- C# Dev Kit extension
Azure Developer CLI
Docker Desktop

What's Included

Awesome Copilot MCP server includes:

Building Block	Name	Description	Usage
Tools	`search_instructions`	Searches custom instructions based on keywords in their descriptions.	`#search_instructions`
Tools	`load_instruction`	Loads a custom instruction from the repository.	`#load_instruction`
Prompts	`get_search_prompt`	Get a prompt for searching copilot instructions.	`/mcp.awesome-copilot.get_search_prompt`

Getting Started

Getting repository root
Running MCP server
Connect MCP server to an MCP host/client
- VS Code + Agent Mode + Local MCP server

Running MCP server

On a local machine

Run the MCP server app using Aspire.
```
aspire run
```

Once running, the Aspire dashboard will be loaded in your default web browser, or you can click the URL provided in the terminal. From here, you'll have access to the MCP server endpoint, logs and metrics.

In a container

Build the MCP server app as a container image.

cd mcp-server
docker build -f Dockerfile -t awesome-copilot:latest .

Run the MCP server app in a container.

docker run -i --rm -p 8080:8080 awesome-copilot:latest

Alternatively, use the container image from the container registry.

docker run -i --rm -p 8080:8080 ghcr.io/github/awesome-copilot:latest

On Azure

Navigate to the directory.
```
cd mcp-server
```

# Login with Azure Developer CLI
azd auth login

Deploy the MCP server app to Azure.
```
azd up
```
While provisioning and deploying, you'll be asked to provide subscription ID, location, environment name.
After the deployment is complete, get the information by running the following commands:
- Azure Container Apps FQDN:
```
azd env get-value AZURE_RESOURCE_MCP_AWESOME_COPILOT_FQDN
```

Connect MCP server to an MCP host/client

Install the MCP server:

Open Command Palette by typing F1 or Ctrl+Shift+P on Windows or Cmd+Shift+P on Mac OS, and search MCP: List Servers.
Choose awesome-copilot then click Start Server.
When prompted, enter one of the following values:
- The absolute directory path of the AwesomeCopilot.McpServer project
- The FQDN of Azure Container Apps.

Use a prompt by typing /mcp.awesome-copilot.get_search_prompt and enter keywords to search. You'll get a prompt like:

Please search all the chatmodes, instructions and prompts that are related to the search keyword, `{keyword}`.

Here's the process to follow:

1. Use the `awesome-copilot` MCP server.
1. Search all chatmodes, instructions, and prompts for the keyword provided.
1. DO NOT load any chatmodes, instructions, or prompts from the MCP server until the user asks to do so.
1. Scan local chatmodes, instructions, and prompts markdown files in `.github/chatmodes`, `.github/instructions`, and `.github/prompts` directories respectively.
1. Compare existing chatmodes, instructions, and prompts with the search results.
1. Provide a structured response in a table format that includes the already exists, mode (chatmodes, instructions or prompts), filename, title and description of each item found. Here's an example of the table format:

    | Exists | Mode         | Filename               | Title         | Description   |
    |--------|--------------|------------------------|---------------|---------------|
    | ✅    | chatmodes    | chatmode1.json         | ChatMode 1    | Description 1 |
    | ❌    | instructions | instruction1.json      | Instruction 1 | Description 1 |
    | ✅    | prompts      | prompt1.json           | Prompt 1      | Description 1 |

    ✅ indicates that the item already exists in this repository, while ❌ indicates that it does not.

1. If any item doesn't exist in the repository, ask which item the user wants to save.
1. If the user wants to save it, save the item in the appropriate directory (`.github/chatmodes`, `.github/instructions`, or `.github/prompts`) using the mode and filename, with NO modification.

Confirm the result.

6.2 KiB Raw Blame History