MCP Leanring

🚀 MCP (Model Context Protocol) Usage Guide

MCP (Model Context Protocol) is a protocol designed to connect clients with Large Language Models (LLMs). This guide will introduce how to use Cline as a client to interact with LLMs such as Gemini.

📚 Reference Links

• MCP Official Documentation
• Zhihu Article

✨ Features

• Connect clients to LLMs like Gemini.
• Support various tools for text processing.
• Define reusable prompt templates and workflows.

📦 Installation

(1) Install the Cline Plugin

Install the Cline plugin in VSCode:

• Open VSCode.
• Go to the plugin marketplace, search for and install Cline.

(2) Configure the MCP Server

The MCP server is mainly used to generate tools (such as those shown in src/calculate.py and src/datawale.py). Here are the configuration steps:

1. Open the VSCode terminal.
2. Run the following command to install dependencies:

   pip install mcp-server
   

3. Create an mcp-config.json configuration file with the following content:

   {
     "llm": {
       "model": "gemini",
       "api_key": "your_api_key"
     },
     "tools": [
       {
         "name": "txt_counter",
         "description": "Counts words in a text file.",
         "path": "src/txt_counter.py"
       }
     ]
   }
   

4. In VSCode, go to the settings interface of the Cline plugin.
5. Select MCP Servers -> Installed, then click Edit Configuration and add the configuration file.

💻 Usage Examples

(1) Run the Server

Ensure that the MCP server is always running for client communication. Run the following command:

/Users/{username}/.local/bin/uv --directory /Users/{username} run txt_counter.py

(2) Run Tools Directly

If uv is not installed, you can directly run the tool script (such as txt_counter.py), but you need to manually modify the port.

(3) Use the Cline Plugin

In VSCode:

1. Open the folder you want to use.
2. Make sure the Cline plugin is installed.
3. Run the following command in the terminal to start the MCP server:

   uv run script.py
   

4. Use the Cline plugin to call tools.

(4) Example Code

Refer to the MCP Official Example:

from mcp.client import Client

client = Client()
response = client.chat("What is 100 + 200?")
print(response)

📚 Documentation

(1) Prompts Structure

Prompts are used to define reusable prompt templates and workflows. Here is its JSON format:

{
  "name": string,              // Unique identifier
  "description": string,      // Description
  "arguments": [
    {
      "name": string,          // Parameter name
      "description": string,   // Parameter description
      "required": boolean     // Whether it is required
    }
  ]
}

(2) Usage Steps

1. Clone the code repository:

   git clone <repo_path>
   cd <repo>
   

2. Install the uv environment:
   • If not installed, run the following command:

     uv install
     

3. Run the script:

   uv run script.py
   

🔧 Technical Details

• The MCP protocol uses the uv server to handle client requests.
• The Cline plugin provides a convenient way to interact with the MCP server in VSCode.
• The mcp-config.json file configures the LLM and available tools.

📄 License

No license information provided.

⚠️ Important Note

• Use absolute paths for commands, as relative paths may cause errors.
• Ensure that the server is always running for clients to call tools.

💡 Usage Tip

• Refer to the official documentation and examples for more detailed information.