Docker Agent is a Docker CLI plugin (docker agent) for defining AI agents in a YAML file and running them without writing code. An agent declaration names a model such as openai/gpt-5-mini, a description, an instruction prompt, and a list of toolsets; docker agent run agent.yaml starts it, and docker agent new generates a configuration interactively.
Agents can be composed into teams of specialized agents that delegate tasks to one another. Tools come from a built-in set plus any Model Context Protocol (MCP) server, whether local, remote, or Docker-based (for example docker:duckduckgo from the MCP catalog). Built-in think, todo, and memory tools support reasoning, and a pluggable RAG layer offers BM25, embeddings, hybrid search, and reranking.
Model providers include OpenAI, Anthropic, Gemini, AWS Bedrock, Mistral, xAI, and Docker Model Runner for local models, with further providers such as DeepSeek, Groq, OpenRouter, Together AI, Hugging Face, and Vercel AI Gateway. Agents are packaged as OCI artifacts: push one to any registry and run it elsewhere with docker agent run myorg/agent:tag. Beyond the CLI, the runtime offers a terminal UI, an MCP mode, A2A and ACP protocol support, an API server, sandbox mode, structured output, hooks, permissions, and a Go SDK.
Features
- YAML agents: declarative, versionable agent definitions with model, instruction, and toolsets
- Multi-agent teams: specialized agents that hand off, transfer, and delegate tasks, including background agents
- Tool ecosystem: built-in file, shell, fetch, and script tools plus any MCP server (local, remote, or Docker-based)
- Reasoning tools: think, plan, todo, memory, and scheduler tools
- RAG: BM25, embeddings, hybrid search, and reranking
- Provider choice: OpenAI, Anthropic, Gemini, Bedrock, Mistral, xAI, Docker Model Runner, and more, with model routing
- OCI distribution: push agents to any registry and pull them anywhere
- Runtime surfaces: terminal UI, headless CLI for CI, MCP mode, A2A and ACP protocols, API and chat servers
