上下文感知 MCP 服务器与 Agent Skills

Context-Aware MCP Server & Agent Skills

An MCP (Model Context Protocol) server that automatically loads relevant markdown context files as guidelines for GitHub Copilot, with support for GitHub Copilot Agent Skills.

Features

• 🎯 Intelligent Context Matching: Automatically selects relevant documentation based on keywords in your prompt
• 📚 Multiple Tools: Load context automatically, list available docs, or fetch specific files
• ⚙️ Configurable: Easy-to-manage manifest.json for adding new context files
• 🔄 Base Context: Always includes foundational guidelines from base.md
• 🖥️ Central Server Mode: Run on a server, connect from multiple clients (see SERVER_SETUP.md)
• ✨ Agent Skills Support: VS Code automatically loads skills from .github/skills (no setup required!)

Setup Options

Option A: Agent Skills (Recommended - Zero Setup!)

VS Code automatically discovers and loads skills from .github/skills/. No configuration needed!

Just commit the .github/skills folder to your repository and Copilot will automatically use them.

👉 See Agent Skills Documentation below for details.

Option B: MCP Server - Local Setup (Each PC)

Run MCP server locally on each machine. Best for individual use or Claude Desktop.

Option C: MCP Server - Central Server Setup (Teams)

Run MCP server on one central server, all PCs connect remotely. No installation needed on client PCs!

👉 See SERVER_SETUP.md for central server instructions

Setup

1. Clone the Repository

git clone https://github.com/713zhao/autoload_context_mcp.git
cd autoload_context_mcp

2. Install Dependencies

pip install -r requirements.txt

3. Configure GitHub Copilot to Use This MCP

For VS Code:

1. Open VS Code Settings (JSON)

   • Press Ctrl+Shift+P (or Cmd+Shift+P on Mac)
   • Type "Preferences: Open User Settings (JSON)"
   • Select it to open your settings.json file

2. Add the MCP Server Configuration

   Add the following configuration to your settings.json (update the path to match where you cloned the repository):

   复制

   {
     "github.copilot.chat.mcpServers": {
       "context-loader": {
         "command": "python",
         "args": ["/absolute/path/to/autoload_context_mcp/mcp_server.py"]
       }
     }
   }
   

   Windows Example:

   复制

   {
     "github.copilot.chat.mcpServers": {
       "context-loader": {
         "command": "python",
         "args": ["c:\\Users\\YourName\\autoload_context_mcp\\mcp_server.py"]
       }
     }
   }
   

   Mac/Linux Example:

   复制

   {
     "github.copilot.chat.mcpServers": {
       "context-loader": {
         "command": "python3",
         "args": ["/home/username/autoload_context_mcp/mcp_server.py"]
       }
     }
   }
   

3. Save the settings.json file

4. Reload VS Code

   • Press Ctrl+Shift+P (or Cmd+Shift+P on Mac)
   • Type "Developer: Reload Window"
   • Press Enter

5. Verify MCP is Active

   • Open Copilot Chat in VS Code
   • The MCP server should now be available as a tool
   • Try asking: "How do I write a mock?" to see it automatically load context

For Claude Desktop:

Add to your claude_desktop_config.json:

Location:

• Windows: %APPDATA%\Claude\claude_desktop_config.json
• Mac: ~/Library/Application Support/Claude/claude_desktop_config.json
• Linux: ~/.config/Claude/claude_desktop_config.json

Configuration:

{
  "mcpServers": {
    "context-loader": {
      "command": "python",
      "args": ["/absolute/path/to/autoload_context_mcp/mcp_server.py"]
    }
  }
}

Restart Claude Desktop after saving.

Usage

Using Agent Skills (Automatic)

When Agent Skills are committed to your repository in .github/skills/, Copilot automatically loads them when relevant:

Ask Copilot: "How do I create a mock class?"
→ Copilot automatically loads gtest-mock skill

Ask Copilot: "How do I run tests with filters?"
→ Copilot automatically loads gtest-execute skill

Ask Copilot: "What design patterns should I use?"
→ Copilot automatically loads design-guidelines skill

No special commands needed! Just ask your question and Copilot intelligently loads the right skill based on your prompt.

Using MCP Server (Manual)

Once configured, you can use these tools in Copilot Chat:

Load Context Automatically

@workspace How do I write a mock for my function?

The MCP will automatically detect "mock" keyword and load relevant testing documentation.

List Available Contexts

@workspace Use load_contexts tool to show what documentation is available

Get Specific File

@workspace Load the Design.md context file

Adding New Context Files

Edit context/manifest.json to add new documentation:

{
  "docs": [
    {
      "path": "context/your-new-doc.md",
      "when": ["keyword1", "keyword2", "trigger-word"]
    }
  ]
}

The when array contains keywords that will trigger this document to be loaded.

File Structure

autoload/
├── mcp_server.py          # Main MCP server
├── context/
│   ├── base.md            # Always-included base context
│   ├── manifest.json      # Configuration for context routing
│   ├── design/
│   │   └── Design.md
│   └── testing/
│       ├── GTest_Mock.md
│       └── GTest_Execute.md
└── agent/                 # Legacy API server (optional)

How It Works

1. When you ask Copilot a question, the MCP server receives your prompt
2. It analyzes keywords in your prompt against the manifest.json configuration
3. It scores each document based on keyword matches
4. It loads the top 3 most relevant documents plus base.md
5. All context is returned to Copilot as guidelines for generating responses

Example

Your prompt: "How do I run specific tests with filters?"

MCP automatically loads:

• base.md (always included)
• GTest_Execute.md (matched "run", "execute", "filter")

The loaded context guides Copilot's response with your specific guidelines and best practices.

Agent Skills (Automatic Loading)

What are Agent Skills?

Agent Skills are an open standard used by GitHub Copilot, Claude, and other AI agents. Skills are automatically discovered and loaded when relevant to your prompt.

Available Skills in This Project

Located in .github/skills/:

1. gtest-mock - Google Mock (gmock) guide

   • Creating mock classes with MOCK_METHOD
   • Setting up EXPECT_CALL expectations
   • Matchers and cardinality
   • Best practices for mocking

2. gtest-execute - Google Test execution and coverage

   • Running tests with filters
   • Achieving 100% CTC coverage
   • Test documentation standards
   • Boundary and input testing strategies

3. design-guidelines - Software design principles

   • Architecture patterns
   • Module structure
   • Dependency injection
   • Best practices and design review checklist

How Agent Skills Work

1. Automatic Discovery: VS Code scans .github/skills/ folders in your repository
2. Intelligent Loading: When you ask a question, Copilot matches keywords in your prompt to skill descriptions
3. Context Injection: Relevant skills are automatically loaded into Copilot's context
4. No Configuration: Works automatically - no settings.json changes needed!

Adding New Agent Skills

1. Create a new subdirectory in .github/skills/

   复制

   mkdir .github/skills/your-skill-name
   

2. Create SKILL.md with YAML frontmatter:

   复制

   ---
   name: your-skill-name
   description: What this skill does and when to use it. Include trigger keywords.
   ---
   
   # Your Skill Content
   
   Documentation, examples, and guidelines here...
   

3. Commit to your repository - Copilot will automatically discover it!

Agent Skills vs MCP Server

Recommendation: Use Agent Skills for new projects - they require zero setup and work automatically!

MCP Server Setup (Alternative Method)