The docker agent serve a2a command starts an HTTP server that exposes your agents using the A2A protocol. This enables communication between Docker Agent and other agent frameworks that implement A2A.
A2A support is functional but still evolving. Tool calls, artifacts, and memory features have limited A2A integration. See the limitations section below.
Usage
# Start an A2A server
docker agent serve a2a ./agent.yaml
# Specify a custom address
docker agent serve a2a ./agent.yaml --listen 127.0.0.1:9000
# Use an agent from the catalog
docker agent serve a2a agentcatalog/pirate
Flags
| Flag | Default | Description |
|---|
-a, --agent <name> | root | Name of the agent to run |
-l, --listen <addr> | 127.0.0.1:8082 | Address to listen on |
--working-dir <path> | (current dir) | Working directory for the session |
How it works
When started, the server:
- Binds to the specified address (or picks an available port if none is given)
- Exposes an A2A-compliant agent card with metadata about the agent
- Accepts incoming A2A requests and routes them to the Docker Agent runtime
- Supports all Docker Agent features — tools, models, model fallbacks — transparently
Connecting to the server
Any A2A-compatible client can connect to the server’s address. The server prints the address on startup:
$ docker agent serve a2a ./agent.yaml
Listening on 127.0.0.1:8082
Current limitations
- Tool calls are handled internally and are not exposed as separate A2A events
- A2A artifact support is not yet integrated
- A2A memory features are not yet integrated
- Multi-agent (sub-agent) scenarios need further work
For exposing agents as MCP tools, see MCP Mode. For stdio-based IDE integration, see ACP. For the HTTP REST API, see API Server.