mcp-code-executor-server

mcp-code-executor-server

0
developer_tools

The MCP Code Executor Server is a robust MCP server designed to enable AI models to execute code across various programming languages in a secure environment. It supports multiple languages and ensures secure execution through isolated containers.

MCP Code Executor Server

A robust Model Context Protocol (MCP) server that enables AI agents to execute code across multiple programming languages in a secure, isolated environment.

🚀 Overview

MCP Code Executor Server standardizes how AI models connect to various programming runtimes by exposing a unified MCP tool endpoint. It allows AI assistants to:

  1. Submit code in a supported language.
  2. Execute it in a controlled, sandboxed environment.
  3. Receive output, including compilation or runtime errors.

🔑 Key Features

  • Multi-Language Support: Java, Python, JavaScript, TypeScript, and C++.
  • Secure Execution: Isolated containers with resource limits (CPU, memory, timeouts).
  • MCP Integration: Conformant MCP server for seamless client discovery.
  • Automatic Compilation: Handles compile-and-run for Java, C++, and TypeScript.
  • Detailed Error Feedback: Returns both compile-time and runtime error messages.
  • Resource Cleanup: Automatically deletes temporary files and enforces execution timeouts.

📐 Architecture

The MCP Code Executor Server follows a client-server model:

  • MCP Client: Protocol client that connects to this server.
  • MCP Server: This application, exposing code-execution tools.
  • Sandboxed Runtimes: Containers or isolated processes per request.

Code Execution Architecture

🏁 Getting Started

Prerequisites

  • Java 17+
  • Python 3.8+
  • Node.js & npm
  • g++ or another C++ compiler

Local Setup

  1. Clone the repository:

    git clone https://github.com/yourusername/mcp-code-executor-server.git
cd mcp-code-executor-server

  2. Build & Run:

    # Build the project
./mvnw clean package

# Start the server
./mvnw spring-boot:run

The server listens on port 8080 by default.

Using Docker

For isolation, run in Docker:

# Build Docker image
docker build -t mcp-code-executor .

# Run container
docker run -p 8080:8080 mcp-code-executor

Or with Docker Compose:

docker-compose up -d

🛠️ Connecting MCP Clients

  1. Start your MCP-compatible AI client.
  2. Point it to http://localhost:8080.
  3. Discover and invoke the code-execution tool.

Example MCP request:

{
  "language": "python",
  "code": "print('Hello, world!')\nprint('hello again')"
}

🤝 Contributing

Contributions are welcome!.

📄 License

Apache License 2.0. See the file for details.