C++ Quick Start MCP Server

This is a Model Context Protocol (MCP) server that equips any compatible AI assistant (Antigravity, Cursor, Claude Desktop, etc.) with the ability to expertly scaffold cross-platform C++ projects.

Instead of writing boilerplate from scratch (and hallucinating build system rules), this MCP server acts as an interactive "Knowledge Base". It guides the AI through a structured interview process and then provides it with exact, proven templates for CMake, Conan, vcpkg, GitHub Actions, and standard directory layouts.

Features

• The go Prompt: Initiates an interactive interview with the developer to determine the exact C++ stack (Standard, Build System, Package Manager, CI/CD, Coding Style, etc.).
• Dynamic Scaffolding Resources: Provides the AI with deep, instructional markdown resources (mcp://scaffold/conan, mcp://scaffold/github-actions, etc.) so it knows exactly how to write the requested boilerplate files perfectly.
• Agent-Ready Architecture: When you run the prompt, your C++ project will be automatically configured for modern agentic development. It seeds your project with built-in guidelines and architecture rules so future AI agents know exactly how to collaborate on your codebase safely and predictably.

Installation

Ensure you have Node.js and npm installed.

Install via NPM (Recommended)

You can install the server directly from the official npm registry:

npm install -g @m-velikov/cpp-quick-start-mcp

Install from Source (Local Development)

If you are cloning this repository to build it locally:

1. Install dependencies:

   npm install

2. Build the server:

   npm run build

3. Install the package globally from the local directory:

   npm install -g .

Once installed via either method, the cpp-quick-start-mcp command will be available system-wide.

Starting the Server (Optional)

If you want to run the server standalone over HTTP/SSE instead of standard I/O:

# Start with default options (port 3000, host 0.0.0.0)
cpp-quick-start-mcp --port 3000

The server binds to all interfaces (0.0.0.0) on port 3000 by default.

Connecting to an AI Agent

This server supports Dual Transport (both stdio for local clients and sse for remote clients).

1. Local Connection (stdio) - DEFAULT

For local IDE clients, you can connect directly without running an HTTP server.

Option A: Global Install (Recommended) If you installed the package globally, use the cpp-quick-start-mcp command directly in the configurations below. (Note: On Windows, use cpp-quick-start-mcp.cmd instead).

Option B: NPX (No Install Required) Alternatively, you can configure your IDE to run the MCP server directly via npx without installing it globally. Simply use npx as the command (or npx.cmd on Windows) and pass -y @m-velikov/cpp-quick-start-mcp as the arguments. For example:

"mcpServers": {
  "cpp-quick-start": {
    "command": "npx",
    "args": ["-y", "@m-velikov/cpp-quick-start-mcp"]
  }
}

Antigravity IDE

1. Open the MCP Store by clicking the "..." dropdown at the top of the editor's side panel or agent panel, and select MCP Servers.
2. Click Manage MCP Servers, then select View raw config to edit your mcp_config.json file.
3. Add the following to your configuration:

   "mcpServers": {
     "cpp-quick-start": {
       "command": "cpp-quick-start-mcp",
       "args": []
     }
   }

   Alternatively, simply ask the Antigravity agent: "Add the cpp-quick-start-mcp server to my configuration".

Codex

Codex uses a TOML file for configuration instead of JSON.

1. Open your ~/.codex/config.toml file (or .codex/config.toml in your project folder).
2. Add the server entry:

   [mcp.servers.cpp-quick-start]
   command = "cpp-quick-start-mcp"
   args = []

   Alternatively, use the Codex CLI:

codex mcp add cpp-quick-start -- cpp-quick-start-mcp

Claude Desktop

Add the server to your Claude Desktop configuration file:

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

{
  "mcpServers": {
    "cpp-quick-start": {
      "type": "stdio",
      "command": "cpp-quick-start-mcp",
      "args": []
    }
  }
}

Restart Claude Desktop after saving.

Claude Code

1. You can add the MCP server directly via the Claude Code CLI using the following command:

   claude mcp add-json cpp-quick-start '{"type":"stdio","command":"cpp-quick-start-mcp","args":[]}' --scope user

2. Or you can manually edit ~/.claude/settings.json and add the JSON block shown above.
3. Reload your editor window or verify active servers with claude mcp list.

GitHub Copilot

1. In Visual Studio Code, open the Command Palette (Ctrl+Shift+P / Cmd+Shift+P) and type MCP: Add Server.
2. Follow the guided flow, using cpp-quick-start-mcp as the command.
3. Alternatively, manually edit your global ~/.copilot/mcp-config.json or project-scoped .mcp/copilot/mcp.json file and add the JSON configuration block.
4. Ensure you are in "Agent mode" in your Copilot chat interface to interact with your added MCP tools.

2. Remote Connection (SSE)

If you start the server using --port, it will automatically switch to HTTP/SSE transport mode. This is extremely useful if you are building a custom MCP client (using the @modelcontextprotocol/sdk client library) or deploying the server to the cloud.

Note: Not all IDEs (like Antigravity or Cursor) currently support configuring remote SSE servers via their mcp.json files yet. For those IDEs, you must use the Local Connection (stdio) method above.

Example configuration for a custom web client:

{
  "mcpServers": {
    "cpp-quick-start": {
      "type": "sse",
      "url": "http://localhost:3000/mcp"
    }
  }
}

Usage

Once you have added the MCP server to your AI client, the agent automatically discovers its capabilities. The server registers a specific prompt (named go) and various instructional resources with the client. When you ask the agent to create or upgrade a C++ project, the agent will scan its available MCP tools and prompts, recognize the cpp-quick-start server as the right tool for C++ scaffolding, and invoke it.

New Projects

Open an empty folder in your editor/terminal and simply ask your AI assistant naturally:

"Help me start a new C++ project"

Or, if your client supports explicit prompt commands, you can trigger the interview directly:

/go help me start a new C++ project

(Depending on your client, this might be /mcp:cpp-quick-start:go)

The AI will take over, interview you about your stack preferences, and generate your custom C++ boilerplate.

Existing Projects

You can also use this MCP server to configure or modernize an existing project for agentic development. Open your existing project folder and tell your AI assistant:

"I want to upgrade this existing C++ project to use modern AI agent best practices."

Or, using the explicit prompt:

/go upgrade the project

The AI will read the exact blueprints from the MCP server and cleanly integrate them into your existing codebase (e.g., adding an AGENTS.md, setting up GitHub Actions, or integrating a package manager).

Community

Got questions, ideas, or need help setting up your C++ environment? Come hang out with us on Discord!

Join the Discord Server