nativeMCP

What is nativeMCP?

nativeMCP is a C++ implementation of the Model Context Protocol (MCP) system, providing a host (CLI-based AI Agent), MCP client, and MCP server framework. It is designed for developers who want to integrate MCP tools with local LLMs (via Ollama) or use MCP servers with other MCP-compatible software like Cursor.

How to use nativeMCP?

Install dependencies (Qt 6.8, cpp-httplib, Ollama), build the project with CMake and Visual Studio 2022, edit config.json to set the Ollama API URL, model, and MCP server definitions, then run the host executable. The host connects to the LLM and automatically calls MCP tools based on the user's natural‑language requests.

Key features of nativeMCP

• C++ implementation of MCP host, client, and server.
• Host acts as an AI Agent with a CLI interface.
• Server base class (MCPServer) for easy tool extension.
• Supports MCP stdio transport only.
• Integrates with Ollama for local LLM inference (e.g., deepseek, qwen).
• Compatible with MCP‑aware applications like Cursor for its servers.

Use cases of nativeMCP

• Build a local AI assistant that can call custom tools (e.g., time, network utilities).
• Develop and test MCP servers written in C++.
• Use nativeMCP servers as tools in external MCP clients (Cursor, etc.).
• Automate multi‑step tasks that combine LLM reasoning with tool execution.

FAQ from nativeMCP

What dependencies does nativeMCP require?

Qt 6.8 (qtbase module), cpp-httplib (via vcpkg), and Ollama with a compatible LLM model (e.g., deepseek, qwen). Build tools: CMake ≥ 3.30 and Visual Studio 2022.

Can I use nativeMCP with an online LLM API?

The current implementation uses a ModelAdapter that communicates with Ollama. The README shows an api_key field in config, which can be left empty for local setups, but it does not confirm support for other online APIs.

What transport protocol does nativeMCP support?

Only stdio transport is implemented for MCP communication.

Can nativeMCP servers be used with Cursor?

Yes, the README states that the server sub‑classes can be used with other MCP‑compatible software like Cursor, using the same configuration format as the official MCP servers.

How do I configure MCP servers in nativeMCP?

Edit the mcpServers section of config.json. On Windows, set command to cmd and args to ["/c"] followed by the interpreter (e.g., the exe path, python, or npx) and server‑specific arguments.