Docker Imploid

docker-imploid

Practical Ephemeral Code with Imploid - A holistic approach to autonomous development and testing.

Overview

This repository provides a secure, containerized environment that integrates two powerful tools into a holistic workflow:

Imploid - Autonomous Code Generation

An AI-powered coding agent built on Claude Code that:

• Works at the conceptual level - You describe what you want, not how to do it
• Handles the full development cycle - From planning to implementation
• Manages git workflows - Creates branches, commits, and PRs automatically
• Maintains context across sessions - Tracks activity and learns from your codebase

Desplega.ai's qa-use MCP - Autonomous Testing & Validation

An MCP (Model Context Protocol) server that enables Claude Code to:

• Generate and execute end-to-end tests - Natural language to executable QA tests
• Visualize test execution - See what the agent is testing in real-time
• Validate implementations automatically - Ensure generated code actually works
• Provide feedback loops - Failed tests inform the next iteration

Why This Combination Matters

As discussed in this article, having imploid generate code is powerful, but having it validate that code works is transformative. This setup provides:

1. Complete autonomy - From feature request to tested implementation
2. Quality assurance - Every change is validated through automated testing
3. Faster iterations - Catch issues immediately, not in production
4. Reduced context switching - No manual testing between code generation cycles

Without qa-use, you'd just have code generation. Without imploid, you'd just have testing. Together, you have a complete autonomous development workflow.

Getting Started

Prerequisites

• Docker & Docker Compose
• SSH keys configured (for git operations)
• API keys:

- Anthropic API key (for Claude) - desplega.ai API key (for qa-use MCP) - GitHub token (optional, for gh CLI operations)

Installation

Based on lessons learned from practical experimentation, here's the battle-tested setup:

1. Clone this repository:

   git clone git@github.com:desplega-ai/docker-imploid.git
   cd docker-imploid

1. Configure your environment:

Create a .env file with your API keys:

   ANTHROPIC_API_KEY=your_anthropic_key
   DESPLEGA_API_KEY=your_desplega_key
   GITHUB_TOKEN=your_github_token  # Optional

1. Set up your project's setup.sh:

Imploid expects a setup.sh file in your project's root directory. This script should: - Install dependencies - Start your application - Ensure it's running on the configured localhost port

Example:

   #!/bin/bash
   npm install
   npm run dev &
   # Wait for server to be ready
   sleep 5

1. Build and start the containers:

   docker-compose up -d

1. Initialize imploid inside the container:

   docker-compose exec imploid bash
   npx imploid@latest --config
   npx imploid@latest --foreground

Key Configuration Details

Following the :

• Non-root user: The container runs as a non-sudo user to safely use --dangerously-skip-permissions
• SSH key mapping: Your ~/.ssh directory is mapped for git operations
• MCP configuration: Claude wrapper automatically loads the MCP config for qa-use integration
• Activity tracking: Map ~/.imploid to persist agent activity and configuration

How It Works

The Complete Workflow

1. You provide a feature request (e.g., "Add user authentication with JWT")

1. Imploid generates the implementation:

- Creates a feature branch - Plans the implementation - Writes the code - Commits changes

1. qa-use validates the implementation:

- Generates end-to-end tests based on the feature - Executes tests against your running application - Visualizes test execution - Reports results back to Claude

1. Feedback loop:

- If tests pass: Creates a PR with confidence - If tests fail: Imploid iterates on the implementation - Process repeats until quality is assured

Using qa-use MCP

The qa-use MCP server enables Claude to:

• Create natural language test scenarios
• Execute them against your running application (localhost:xxxx)
• Capture screenshots and interaction flows
• Provide detailed test reports
• You can later promote your tests to a regression test suite in production or staging.

The MCP handles all the complexity of browser automation, test execution, and result reporting.

License

This project is licensed under the MIT License - see the LICENSE file for details.