Docker Extensions for Automation

Build and manage AI agents, MCP servers, LLM models, and infrastructure stacks — all from Docker Desktop.

A suite of 6 Docker Desktop extensions + ready-to-use examples for turning Docker into a full automation control plane.

---

Extensions

Extension	What It Does	Backend	Status
Stack Commander	Manage all Compose stacks from one UI — start, stop, restart, logs	Go	Scaffolded
Agent Orchestrator	Visual AI agent lifecycle management — health, tasks, resources	Go	Scaffolded
MCP Gateway Manager	Configure MCP Gateway, register MCP servers, browse tools	Go	Scaffolded
Model Runner Dashboard	Native LLM serving via Compose models: key — pull, monitor, compare vs Ollama	Go	Scaffolded
Port Dashboard	Real-time port mapper with conflict detection and one-click open	SDK-only	Scaffolded
Health Monitor	Container health tracking, auto-restart policies, resource alerts	Go	Scaffolded

Examples

Example	What It Demonstrates
MCP Gateway	Secure MCP tool access for containerized agents via docker/mcp-gateway:latest
Compose Models	Native LLM serving with the models: Compose key — zero Ollama needed

---

Architecture

                        Docker Desktop
  ┌──────────┬──────────┬──────────┬──────────┬─────────┬────────┐
  │  Stack   │  Agent   │   MCP    │  Model   │  Port   │ Health │
  │Commander │Orchestr. │ Gateway  │  Runner  │Dashboard│Monitor │
  ├──────────┴──────────┴──────────┴──────────┴─────────┴────────┤
  │                  Docker Extensions SDK                        │
  │    listContainers | cli.exec | vm.service | events stream     │
  ├───────────────────────────────────────────────────────────────┤
  │                  Docker Engine + Compose                      │
  ├──────────────┬───────────────┬────────────────┬───────────────┤
  │ Compose      │  MCP Gateway  │  Model Runner  │    Agent      │
  │ Stacks       │  (tool broker)│  (LLM serving) │  Containers   │
  └──────────────┴───────────────┴────────────────┴───────────────┘

Each extension is independent — install one or all. No cross-dependencies.

---

Quick Start

Prerequisites

• Docker Desktop 4.43.0+ (for Model Runner and models: key)
• Docker Extensions SDK enabled
• Go 1.22+ and Node.js 20+ (for building from source)

Install an Extension Locally

# Build and install Stack Commander
cd extensions/stack-commander
docker build -t tiranmoskovitch/stack-commander .
docker extension install tiranmoskovitch/stack-commander

Run an Example

# Try the MCP Gateway pattern
cd examples/mcp-gateway
docker compose up --build

# Try native model serving
cd examples/compose-models
docker compose up

Enable Extension Development Mode

In Docker Desktop: Settings > Extensions > uncheck "Allow only extensions distributed through the Docker Marketplace"

---

MCP Gateway Pattern

The MCP Gateway example shows how to give AI agents secure access to tools without host access:

services:
  mcp-gateway:
    image: docker/mcp-gateway:latest
    use_api_socket: true
    command:
      - --transport=sse
      - --servers=postgres,filesystem
      - --tools=query,read_file

  agent:
    environment:
      - MCP_SERVER_URL=http://mcp-gateway:8811/sse

Why this matters:

• Agents only see the gateway endpoint — no Docker socket, no host filesystem
• Gateway controls exactly which tools are exposed
• Secrets stay in Docker secrets, not container env vars
• Works with any MCP server that uses stdio transport

---

Compose Native Models

The Compose Models example demonstrates Docker's built-in LLM serving:

# Top-level key — Docker handles everything
models:
  qwen3:
    model: ai/qwen3:14B-Q6_K

services:
  agent:
    # MODEL_RUNNER_URL and MODEL_RUNNER_MODEL are auto-injected
    environment:
      - MODEL_RUNNER_URL
      - MODEL_RUNNER_MODEL

What Docker does automatically:

1. Pulls the model on first compose up
2. Serves it via OpenAI-compatible API
3. Injects endpoint URL into your services
4. Manages GPU allocation

vs. Ollama: Compose models are declarative and zero-config. Keep Ollama if you need broader Windows GPU support or Ollama-specific features today.

---

Project Structure

docker-extensions-automation/
├── extensions/
│   ├── stack-commander/          # P0 — Compose stack management
│   │   ├── metadata.json
│   │   ├── Dockerfile
│   │   ├── docker-compose.yaml
│   │   ├── backend/              # Go backend
│   │   └── ui/                   # React frontend
│   ├── agent-orchestrator/       # P1 — AI agent lifecycle
│   ├── mcp-gateway-manager/      # P1 — MCP server management
│   ├── model-runner-dashboard/   # P1 — Native LLM management
│   ├── port-dashboard/           # P2 — Port mapping & health
│   └── health-monitor/           # P2 — Container health & alerts
├── examples/
│   ├── mcp-gateway/              # MCP Gateway pattern demo
│   └── compose-models/           # Native model serving demo
├── shared/
│   └── sdk-utils/                # Shared React hooks for SDK
├── docs/
│   ├── PROJECT_PLAN.md           # Roadmap and milestones
│   └── ARCHITECTURE.md           # Technical architecture
├── LICENSE                       # MIT
└── README.md                     # This file

Tech Stack

• Frontend: React 18 + TypeScript + Tailwind CSS
• Backend: Go 1.22 (where persistent state is needed)
• SDK: @docker/extension-api-client v0.3.4
• Build: Multi-stage Dockerfiles
• Distribution: Docker Hub under tiranmoskovitch/ namespace

Roadmap

Phase	What	Status
1	Stack Commander MVP	In Progress
2	Agent Orchestrator + MCP Gateway + Model Runner	Planned
3	Port Dashboard + Health Monitor	Planned
4	Docker Hub marketplace publish	Planned

Contributing

PRs welcome. Each extension is independent — you can contribute to one without touching others.

License

MIT