Docker has released a new command-line plugin that enables users to create, configure, and execute AI agents through declarative configuration files. The plugin, distributed as a Docker CLI extension, allows agents to be defined in YAML format, assigned specific capabilities, and coordinated within multi-agent systems.

The tool functions as a Docker CLI plugin and can be invoked via the 'docker agent' command. Agents are configured through YAML files that specify model providers, operational instructions, and tool access. For example, a basic configuration might designate an OpenAI GPT-5-mini model with DuckDuckGo search capabilities as a default assistant.

Key capabilities include:

- Multi-agent orchestration allowing specialized agents to delegate tasks

- Integration with Model Context Protocol (MCP) servers for extended functionality

- Support for multiple AI providers including OpenAI, Anthropic, Google, and Mistral

- Built-in tools for reasoning, task management, and memory

- Retrieval-augmented generation with multiple search methodologies

- Distribution through OCI container registries

Installation options include Docker Desktop (version 4.63 or later), Homebrew, or manual binary deployment. Users must provide API credentials for cloud-based models or configure local model execution through Docker Model Runner. The plugin collects anonymous usage metrics for product improvement.

Example workflows involve running predefined agents, building custom configurations interactively, or executing agents from container registries. The project maintains documentation covering setup procedures, model configuration, and advanced features like retrieval systems and multi-agent architectures.

Community engagement occurs through Docker's Slack community and dedicated '#docker-agent' channel. The plugin represents Docker's expansion beyond container management into AI workflow orchestration, offering a standardized approach to agent-based computing.