A multi-agent CLI orchestrator via the ACP (Agent Client Protocol). Manage 20+ coding agents from one terminal.
Every coding agent has its own CLI, its own workflow, its own way of doing things. agent-team gives them a single control plane:
- One interface for all agents: Prompt, cancel, approve permissions, configure — the same commands work across 20+ agents
- Independent sessions: Each agent runs in its own process with its own UDS socket. No shared state, no interference
- Any terminal, any agent: Send prompts, review permissions, read logs — all from wherever you are
npm install -g agent-teamUpdate to the latest version:
agent-team update# Start a Gemini agent (foreground, Ctrl+C to exit)
agent-team add gemini
# Start a Claude agent in the background
agent-team add claude -b
# List running agents
agent-team ls
# Send a prompt
agent-team ask gemini-1 "Refactor the auth module"
# Read the conversation
agent-team log gemini-1All agents listed on agentclientprotocol.com:
gemini copilot goose claude* codex* auggie kiro cline blackbox openhands qoder opencode kimi vibe qwen cagent fast-agent stakpak vtcode pi*
* Requires a separate adapter binary. agent-team will prompt you with install instructions if not found in PATH.
| Command | Description |
|---|---|
add <type> |
Start agent session (foreground). -b for background |
rm <name> |
Shut down agent. --all for all agents |
ls |
List running agents |
restart <name> |
Restart agent (preserves config) |
info <name> |
Show agent details |
| Command | Description |
|---|---|
ask <name> [text] |
Send prompt and wait for response. -f to attach files |
log <name> |
Read conversation. -n N for last N messages, -a for agent-only |
cancel <name> |
Cancel current task |
allow/deny <name> |
Approve or reject permission request |
| Command | Description |
|---|---|
mode <name> <mode> |
Switch agent mode (ask/code/architect) |
set <name> <key> <val> |
Change runtime config |
update |
Self-update via npm |
The simplest approach - just tell your agent to use it:
Use agent-team to manage coding agents. Run agent-team --help for usage.
The --help output is comprehensive and most agents can figure it out from there.
For more consistent results, add to your project instructions:
## Multi-Agent Orchestration
Use `agent-team` to spin up and coordinate coding agents. Run `agent-team --help` for all options.
Usage:
- `agent-team add gemini -b` - Start a Gemini agent in background
- `agent-team ls` - List running agents
- `agent-team ask <name> "task"` - Send a prompt and wait for response
- `agent-team log <name> -a -n 1` - Read last agent response
- `agent-team cancel <name>` - Cancel current task
- `agent-team allow/deny <name>` - Approve or reject permission request
- `agent-team rm <name>` - Shut down agent
Typical workflow:
1. Start agents: `agent-team add gemini -b && agent-team add copilot -b`
2. Assign tasks: `agent-team ask gemini-1 "Refactor auth module"`
3. Check progress: `agent-team log gemini-1`
4. Clean up: `agent-team rm --all`MIT