Code Executor MCP Server

By MEOK AI Labs — Sovereign AI tools for everyone.

Code Executor MCP Server

Sandboxed code execution for AI agents. Run Python, JavaScript, and shell commands with comprehensive safety guards, output capture, timeout protection, and file access restrictions. Built for production use with blocked destructive patterns and configurable limits.

Tools

Safety Features

• Blocks destructive shell commands: rm -rf /, fork bombs, pipe-to-shell
• Blocks dangerous Python: os.system, subprocess, eval(input), raw sockets
• Blocks dangerous JavaScript: child_process, require('fs'), eval()
• File access restricted to Desktop, Documents, Downloads, /tmp
• All execution happens in an isolated temp directory
• Hard timeout cap at 60 seconds
• Output truncated at 10KB to prevent memory issues

Installation

pip install mcp

Optional for JavaScript support:

brew install node  # or: apt install nodejs

Usage

Run the server

python server.py

Claude Desktop config

{
  "mcpServers": {
    "code-executor": {
      "command": "python",
      "args": ["/path/to/code-executor-mcp/server.py"]
    }
  }
}

Example calls

Execute Python:

Tool: execute_code
Input: {"code": "import math\nprint(f'Pi = {math.pi:.10f}')\nprint(f'e = {math.e:.10f}')", "language": "python"}
Output: {"output": "Pi = 3.1415926536\ne = 2.7182818285\n", "exit_code": 0, "elapsed_seconds": 0.045}

Run shell command:

Tool: run_command
Input: {"command": "ls -la /tmp | head -20"}
Output: {"output": "total 128\ndrwxrwxrwt ...", "exit_code": 0, "elapsed_seconds": 0.012}

Run tests:

Tool: run_tests
Input: {"test_command": "python -m pytest tests/ -v", "working_dir": "/path/to/project"}
Output: {"passed": true, "summary": "12 passed in 1.34s", "exit_code": 0}

Blocked command example:

Tool: run_command
Input: {"command": "rm -rf /"}
Output: {"error": "Command blocked by safety filter (matches: rm\\s+-rf\\s+/)"}

Pricing

License

MIT