# cli-agent-orchestrator [Health: Active]

**Category:** 🤖 Coding Agents  
**Repository:** https://github.com/awslabs/cli-agent-orchestrator  
**GitHub Stars:** 1485  
**Views:** 0  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/cli-agent-orchestrator

## Description
Multi-agent orchestration for AI coding CLIs — Claude Code, Kiro, Codex, and more, coordinated in isolated tmux sessions

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `uvx` (confidence: high):

```json
"mcpServers": {
  "cli-agent-orchestrator": {
    "command": "uvx",
    "args": ["--from","cli-agent-orchestrator","cao-mcp-server"]
  }
}
```

## Documentation

## What cli-agent-orchestrator does

cli-agent-orchestrator coordinates AI coding command-line tools under the direction of a supervisor. Instead of replacing each provider's CLI, it starts those CLIs as ordinary processes and preserves their native authentication and capabilities. Workers run in isolated tmux sessions, allowing a supervisor to assign work to multiple agents at once or arrange tasks into a sequence.

The project fits the Coding Agents category and provides several control surfaces around the local orchestration runtime, including a shell CLI, a web interface, an operations MCP server, and plugins. It also includes support for flows, workflows, agent profiles, skills, memory, and optional self-learning features.

## How it works

The runtime is built around a local `cao-server`. A typical session has the server running in one terminal, while a second terminal launches a supervisor from the project directory. CAO then starts the selected provider CLIs in separate tmux sessions. Operators can watch the attached launch terminal, use the web UI at `http://localhost:9889`, or attach to the relevant tmux session.

The built-in `code_supervisor` profile provides a starting point for supervisor-driven work. Sessions can be stopped individually by name or shut down together with the all-sessions command. Agent profiles can select providers and define overrides, while tool restrictions can apply roles, allowlists, and provider enforcement.

## Setup and configuration

CAO requires Python 3.11 or newer, tmux 3.3 or newer, and `uv`. At least one supported provider CLI must be installed and authenticated before CAO starts. Providers listed by the project include Kiro CLI, Claude Code, Codex CLI, Antigravity CLI, Hermes, Kimi CLI, MiniMax Code, GitHub Copilot CLI, OpenCode CLI, Oh My Pi, Cursor CLI, and Grok Build CLI.

Install the main branch as a `uv` tool with:

```bash
uv tool install git+https://github.com/awslabs/cli-agent-orchestrator.git@main --upgrade
```

After installation, install the supervisor profile, start the server, and launch it from the target project:

```bash
cao install code_supervisor
cao-server
cao launch --agents code_supervisor
```

Provider-specific guides cover installation, authentication, and provider overrides. CAO can also be installed from PyPI or through its devcontainer feature. The `cao update` command updates an existing installation.

## Tools and capabilities

The supervisor can coordinate workers in parallel or sequence and can use the operations MCP server as a control plane. Other available control surfaces include:

- A shell CLI for launching, updating, and shutting down sessions
- A web UI for observing and operating the agent fleet
- MCP Apps for browser or host-rendered interfaces
- Flows and workflows for scheduled or multi-step runs
- Reusable skills with installation, discovery, loading, updating, and removal
- Persistent cross-session memory and an opt-in self-learning loop
- Agent and event plugin support, with agent plugins gated off by default
- HTTP API and PTY WebSocket routes for API and terminal streaming use cases

_Full upstream README: https://allmcps.com/mcp/cli-agent-orchestrator/readme_