CPP MCP-SERVER

A C++ implementation of a Model Context Protocol Server with a pluggable module architecture.

* Resource Templates aren't supported yet

** Partially supported

Claude Code Configuration

claude mcp add --transport http cpp_mcp_server http://localhost:8080/mcp

Setup Logs Directory

By default, the mcp server logs into ./logs directory (MUST BE EXISTS) You can redirect logs into your preferred dir using the -l command line option.

Supported Transports

Supported Platforms

How to compile

Clone the repository on your local machine

git clone https://github.com/peppemas/mcp_server.git

Compile

cd mcp_server
mkdir build
cd build
cmake ..
make

MCP Server Architecture

The MCP Server is designed to implement a Model Context Protocol, enabling a modular and extensible architecture. Below is an overview of the project's architecture:

Core Components

1. Server Core:
   The central component of the server that handles the initialization and management of all major functionalities, including communication with clients, logging, and plugin management.

2. Command-Line Interface:
   Provides command-line arguments for configuration:

   • -n: Specifies the name of the server to expose to clients. (optional)
   • -p: Sets the path to the plugins directory where to search the plugins (it searches also in subdirectories).
   • -l: Sets the path to the log directory.

3. Plugin System:
   The server is designed to load plugins dynamically from a specified directory (-p argument). Each plugin extends the server's functionality, allowing for future expansion without modifying the core codebase.

4. Logging Module:
   Logging is centralized, with logs stored in the directory specified by the -l parameter. This enhances debugging and monitoring of the server's activity.

Execution Flow

1. Initialization:

   • The server starts by parsing the command-line arguments to configure the server name, plugin directory, and logs directory.
   • Logs are initialized, and the plugin system is prepared.

2. Plugin Loading:

   • Plugins from the specified directory are identified and loaded dynamically.
   • Each plugin is initialized and registered with the server.

3. Handling Client Requests:

   • The server listens for incoming client connections.
   • It processes requests based on the implemented Model Context Protocol, which relies on registered plugins for extended capabilities.

4. Error Handling:

   • Errors during startup or runtime are logged in the logging directory, ensuring minimal disruption to the server's operation.

Plugin Extensibility

The MCP Server's extensibility is powered by its plugin system, enabling developers to expand the server's capabilities without modifying the core logic. Plugins are dynamically loaded libraries and can be implemented in various ways depending on the operating system:

• On Windows: Plugins are compiled as .dll files.
• On Linux/MacOS: Plugins are compiled as .so or .dylib (shared object) files.

Example Plugins

1. Weather Plugin:
   The Weather plugin allows the server to provide weather-related functionalities. This plugin can process requests related to weather data, such as current temperature, forecasts, and other meteorological information. The plugin fetches data from third-party APIs to ensure up-to-date and accurate information. It demonstrates how the MCP server can be extended to offer services that require external data integration.

2. Sleep Plugin:
   The Sleep plugin introduces delay or "sleep" functionality within the server's processing pipeline. This can be useful for simulating response delays or managing timed operations in protocols. Developers can leverage this plugin to build and test features related to time-based operations without altering core server behavior. It highlights how custom plugins can target specific, niche use cases.

3. Code Review Plugin:
   The Code Review plugin is designed to assist developers in performing automated code reviews. This plugin analyzes code submitted to the server and provides helpful feedback, such as identifying potential bugs, code smells, and optimization opportunities. It follows standard coding practices and can be configured to enforce specific style guides (e.g., Google C++ Style Guide). Additionally, it supports multi-language inspection, ensuring compatibility with various programming languages. The plugin can integrate with version control systems to retrieve code changes directly for assessment, making it a valuable tool for collaborative development workflows.

Claude Desktop Configuration

{
  "mcpServers": {
    "mcp-server": {
      "command": "C:\\mcp-server\\mcp_server.exe",
      "args": [
        "-n","developer-server",
        "-l","C:\\mcp-server\\logs",
        "-p","C:\\mcp-server\\plugins"
      ],
      "env": {
        "CUSTOM_API_KEY_1": "your-api-key-here",  
        "CUSTOM_API_KEY_2": "your-api-key-here",
        "SAVE_DIR": "/path/to/save/directory"
      }      
    }
  }
}

NOTE: In the "env" block you can pass parameters to your plugins e

TODO LIST

• Review log level implementation
• Add json schema validator (https://github.com/pboettch/json-schema-validator)
• Test on different mcp-client (actually tested only on Claude Desktop)