Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/docker/docker-agent/llms.txt

Use this file to discover all available pages before exploring further.

The docker agent serve mcp command makes your agents available to any application that supports the Model Context Protocol. This lets you:
  • Use custom agents directly within Claude Desktop or Claude Code
  • Share specialized agents across different applications
  • Build reusable agent teams consumable from any MCP client
  • Integrate domain-specific agents into existing workflows

​
Usage

# Expose a local config via stdio (default)
docker agent serve mcp ./agent.yaml

# Expose from the agent catalog
docker agent serve mcp agentcatalog/pirate

# Set the working directory
docker agent serve mcp ./agent.yaml --working-dir /path/to/project

# Use streaming HTTP transport instead of stdio
docker agent serve mcp ./agent.yaml --http --listen 127.0.0.1:9090

​
Flags

FlagDefaultDescription
-a, --agent <name>(all agents)Name of the agent to expose
--httpfalseUse streaming HTTP transport instead of stdio
-l, --listen <addr>127.0.0.1:8081Address to listen on (HTTP transport only)
--working-dir <path>(current dir)Working directory for the session

​
How it works

By default, serve mcp uses stdio transport: the MCP client spawns the docker agent serve mcp process as a subprocess and communicates via stdin/stdout. This requires no network configuration. With --http, it starts a streaming HTTP server that MCP clients can connect to over the network. Each agent in your config file becomes a separate tool in the MCP client. For multi-agent configs, all agents are exposed.

​
Setting up Claude Desktop

Add a server entry to your Claude Desktop MCP configuration file:
  • macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
  • Windows: %APPDATA%\Claude\claude_desktop_config.json
{
  "mcpServers": {
    "myagent": {
      "command": "/usr/local/bin/docker",
      "args": [
        "agent",
        "serve",
        "mcp",
        "agentcatalog/coder",
        "--working-dir",
        "/home/user/projects"
      ],
      "env": {
        "ANTHROPIC_API_KEY": "your_key_here",
        "OPENAI_API_KEY": "your_key_here"
      }
    }
  }
}
Restart Claude Desktop after saving.

​
Setting up Claude Code

claude mcp add --transport stdio myagent \
  --env OPENAI_API_KEY=$OPENAI_API_KEY \
  --env ANTHROPIC_API_KEY=$ANTHROPIC_API_KEY \
  -- docker agent serve mcp agentcatalog/pirate --working-dir $(pwd)

​
Multi-agent configs

When you expose a multi-agent configuration, each agent becomes a separate tool: 
agents:
  root:
    model: anthropic/claude-sonnet-4-0
    description: Main coordinator
    sub_agents: [designer, engineer]
  designer:
    model: openai/gpt-4o-mini
    description: UI/UX design specialist
  engineer:
    model: anthropic/claude-sonnet-4-0
    description: Software engineer
All three agents (root, designer, engineer) appear as separate tools in the MCP client.

​
Troubleshooting

ProblemSolution
Agents not appearingVerify the docker binary path and restart the MCP client
Permission errorsEnsure the docker binary has execute permissions
Missing API keysPass all required keys in the env section
Working directory issuesVerify the --working-dir path exists

Build docs developers (and LLMs) love

Get started for freeTalk to us