CPPCon MCP Server

This project implements a Model Context Protocol (MCP) server that provides a tool to compile C++ code using the Compiler Explorer (godbolt.org) API. It allows clients to request the assembly output for given C++ code and compiler options, supporting both standard tool invocation and Server-Sent Events (SSE) for streaming results.

Features

• MCP Compliant: Built using mcp-server[fastmcp], ensuring adherence to the Model Context Protocol.

• get_assembly Tool: Exposes a tool that compiles C++ code using MSVC (via godbolt.org) and returns the generated assembly.

• SSE Support: Provides a Server-Sent Events endpoint (/sse) for streaming compilation status and results.

• Tool Discovery: Exposes a /capabilities endpoint for clients to discover available tools and their schemas.

• Client Messaging: Supports a /message endpoint for general client-to-server communication.

Setup

Prerequisites

• Python 3.11 or higher

Installation

1. Create a virtual environment (recommended):

   python -m venv venv
   source venv/bin/activate  # On Windows, use `venv\Scripts\activate`

2. Install dependencies:

   pip install "mcp-server[fastmcp]" uvicorn requests pydantic

Running the Server

To start the MCP server, execute the server.py script:

python server.py

The server will run on http://127.0.0.1:8000 by default.

Usage

The server exposes several endpoints compliant with the Model Context Protocol.

1. Get Server Capabilities

Clients can discover the server's capabilities, including available tools, by sending a GET request to the /capabilities endpoint.

curl http://127.0.0.1:8000/capabilities

2. Invoke get_assembly Tool (Standard POST)

To get the assembly output for C++ code in a single request/response cycle, send a POST request to the /tool/get_assembly endpoint.

Example Request:

curl -X POST 
     -H "Content-Type: application/json" 
     -d '{"code": "int main() { return 0; }", "options": "/O2"}' 
     http://127.0.0.1:8000/tool/get_assembly

3. Stream get_assembly Results (Server-Sent Events)

For streaming compilation status and results, connect to the /sse endpoint using GET. The code and options for compilation are passed as URL query parameters. An optional sessionId can also be provided for client tracking.

Example Request:

curl -N "http://127.0.0.1:8000/sse?code=int%20main()%20%7B%20return%200%3B%20%7D&options=/O2&sessionId=my_compiler_session_123"

The server will stream events in the following format:

event: message
data: {"status": "starting_compilation", "sessionId": "my_compiler_session_123"}

event: result
data: {"type": "text", "text": "; Assembly output here..."}

event: message
data: {"status": "compilation_complete", "sessionId": "my_compiler_session_123"}

In case of an error, an event: error will be sent:

event: error
data: {"type": "server_error", "text": "Error details here...", "sessionId": "my_compiler_session_123"}

4. Send Client Messages

Clients can send general messages to the server via a POST request to the /message endpoint. These messages can be used for various client-initiated communications.

Example Request:

curl -X POST 
     -H "Content-Type: application/json" 
     -d '{"type": "client_ping", "sessionId": "my_compiler_session_123", "data": "Are you alive?"}' 
     http://127.0.0.1:8000/message

The server will respond with:

{"status": "message_received", "sessionId": "my_compiler_session_123"}

5. Test Server Using modelcontextprotocol/inspector

npx @modelcontextprotocol/inspector

6. Configure using VSCode http MCP server command create mcp.json and start tool

{
    "servers": {
        "my-mcp-server-4620e938": {
            "url": "http://localhost:8000/sse",
            "type": "http"
        }
    },
    "inputs": []
}

7. Chat with Copilot, when context need assembly, it will invoke