Onyx

avd1729/Onyx

3.2

If you are the rightful owner of Onyx and would like to certify it and/or have it hosted online, please leave a comment on the right or send an email to henry@mcphub.com.

Onyx is a Model Context Protocol (MCP) server designed to execute code securely within Docker sandboxes, supporting multiple programming languages and integrating with Claude Desktop or other MCP clients.

Tools
1
Resources
0
Prompts
0

Onyx

Onyx is a MCP (Model Context Protocol) server that executes code securely inside Docker sandboxes. It supports multiple programming languages and integrates seamlessly with Claude Desktop or other MCP clients. Onyx makes it safe and easy to execute arbitrary code in a sandboxed MCP environment. Perfect for Claude Desktop power users or anyone building AI workflows with executable code.

Features

  • πŸ”Ή Multi-language support:

    • Python (python:3.11)
    • Java (openjdk:17)
    • C (gcc:12)
    • C++ (gcc:12)
    • JavaScript / Node.js (node:20)
    • Rust (rust:1.72)
  • πŸ”Ή Docker sandboxing:

    • Network disabled (--network none)
    • Read-only FS with tmpfs mounts for safe writes
    • Limited CPU, memory, and process count
    • Non-root execution (--user 1000:1000)
  • πŸ”Ή MCP protocol: Exposes a run_code tool for executing arbitrary code.

  • πŸ”Ή Logging: All execution logs go to stderr (safe for MCP clients).

  • πŸ”Ή CI-ready: Automated tests for executors via GitHub Actions.


Project Structure

sandbox/
β”œβ”€β”€ .github/
β”‚   └── workflows/ci.yml       # GitHub Actions CI
β”œβ”€β”€ cmd/
β”‚   └── server/
β”‚       └── main.go            # MCP server entrypoint
β”œβ”€β”€ internal/
β”‚   β”œβ”€β”€ model/                 # Code params, executor interface, result types
β”‚   β”œβ”€β”€ executor/              # Language-specific executors
β”‚   └── utils/                 # Docker availability checks
β”œβ”€β”€ tests/
β”œβ”€β”€ go.mod, go.sum

Setup Instructions

1. Prerequisites

Verify both:

go version
docker --version

2. Pull Required Docker Images

To avoid delays during the first execution, pre-pull all language runtimes:

docker pull python:3.11
docker pull openjdk:17
docker pull gcc:12
docker pull node:20
docker pull rust:1.72

3. Build and Run the Server

From the root of the repo:

# Run directly
go run ./cmd/server/main.go

# Or build a binary (recommended for Claude Desktop)
go build -o sandbox_server ./cmd/server/main.go

You should now have sandbox_server (or sandbox_server.exe on Windows).


Connecting with Claude Desktop

  1. Locate Claude Desktop’s config file:

    code $env:AppData\Claude\claude_desktop_config.json
    
  2. Add an MCP server entry for Onyx:

    {
      "mcpServers": {
        "onyx": {
          "command": "<absolute path>/sandbox_server.exe",
          "args": []
        }
      }
    }
    
  3. Restart Claude Desktop.

  4. You should now see the run_code tool available.


Usage

image For a good example of how it can be used, check out this chat: https://claude.ai/share/512e1f5a-77fc-40ca-8c49-b372b2c680f9

Extending to New Languages

Adding support for another language requires:

  1. Choose a Docker image with the language’s compiler/runtime. Example: golang:1.22 for Go, php:8.2-cli for PHP.

  2. Implement an Executor in internal/executor/<lang>.go:

    • Define a struct (e.g., GoExecutor).

    • Implement the Execute method to:

      • Pipe source code into the container.
      • Compile/interpret it inside /workspace.
      • Run the output binary/script.
  3. Register it in main.go:

    if args.Language == "go" {
        runtime = executor.GoExecutor{}
    }
    
  4. Write a test in internal/executor/go_test.go.

  5. Pull the Docker image ahead of time (docker pull golang:1.22).