Docker Agent Tutorial: Run Your First Multi-Agent Team in Minutes

In this hands-on tutorial you will go from zero to a running multi-agent team with Docker Agent — no application code required. If you have Docker Desktop 4.63+ the plugin is already installed: just run docker agent. Homebrew users can brew install docker-agent.

Step 1: Set up a model

Export at least one API key, or use Docker Model Runner for fully local inference:

export OPENAI_API_KEY=sk-...   # or ANTHROPIC_API_KEY, GOOGLE_API_KEY

Step 2: Define your first agent

Create agent.yaml with a single assistant and give it a web-search tool via MCP:

agents:
  root:
    model: openai/gpt-5-mini
    description: A helpful AI assistant
    instruction: |
      You are a knowledgeable assistant that helps users with various tasks.
      Be helpful, accurate, and concise.
    toolsets:
      - type: mcp
        ref: docker:duckduckgo

Step 3: Run it

Run your agent with one command, or pull a ready-made agent straight from a registry:

docker agent run agent.yaml
docker agent run myorg/agent:tag
docker agent new   # interactive generator

Step 4: Scale to a team

Add more agents to the YAML and Docker Agent orchestrates them: a researcher agent can gather facts, a writer agent can draft, and a reviewer agent can polish — with automatic delegation between them. Built-in think, todo, and memory tools keep complex tasks on track, and RAG retrievers (BM25, embeddings, hybrid, reranking) ground answers in your own data.

Review

Verdict: Docker Agent is one of the quickest ways to go from idea to running AI agents, and its OCI-based sharing model makes agents as portable as containers. Perfect for teams that already live in Docker.