# Docker Agent

> Apache-2.0 YAML-defined agents with MCP toolsets — runs as `docker agent`

**Category:** AI Agents  
**Pricing:** open-source  
**URL:** https://docker.github.io/docker-agent/  
**GitHub:** https://github.com/docker/docker-agent  
**Added:** 2026-10-08

---

## The Hook

Docker Agent (formerly cagent) lets you declare an agent — model, instructions, MCP toolsets, sub-agents — in one YAML file and run it with `docker agent run agent.yaml`, so agents version and share like a Compose file.

## What's Good

Apache-2.0 (docker/docker-agent; the old docker/cagent URL redirects). Release v1.149.0 (2026-10-07). Pre-installed as a CLI plugin in Docker Desktop 4.63+, or `brew install docker-agent`. Built-in tools plus any MCP server (local, remote, or Docker-based), multi-agent orchestration, TUI, MCP mode, and RAG. Provider-agnostic: OpenAI, Anthropic, Gemini, Bedrock, Mistral, xAI, or local models via Docker Model Runner. ★~3.8k.

## The Catch

Agents are YAML config, not code in an SDK — great when your tools already ship as Docker images or MCP servers, limiting when you need custom control flow in Python/TypeScript. Local models need Docker Model Runner; otherwise bring your own provider API key and pay per token. Vs crewai (Python role/crew framework) and strands-agents (AWS model-driven SDK): Docker Agent is the declarative, container-native lane.

## Verdict

Apache-2.0 declarative agent runner when your tools already live in Docker/MCP — YAML over SDK code; BYO key or Docker Model Runner.