Multi-Model CLI Orchestrator (MMCO)
Use the right AI for the job. A Claude Code plugin that intelligently delegates tasks to specialist LLM CLIs based on their strengths.
Different AI tools excel at different tasks. Kimi nails frontend design. Gemini crushes research. Codex handles massive codebases. But switching between them is a pain.
MMCO fixes this. It analyzes your task and automatically routes it to the best tool—or lets you explicitly delegate with a simple @cli command.
You: @kimi design a modern dashboard with dark mode
───────────────────────────────────────────
kimi Creating Dashboard.tsx...
───────────────────────────────────────────
[kimi] import { LineChart } from 'recharts';
[kimi] export function Dashboard() { ...
───────────────────────────────────────────
kimi ✓ Complete · 3 files · 2.3s
───────────────────────────────────────────
Features
Intelligent Routing
Tasks are analyzed and routed based on confidence scores:
| Confidence | Action | Example |
|---|
| ≥80% | Auto-delegate | "Design a landing page" → Kimi |
| 50-80% | Suggest | "Create a nice UI" → Suggest Kimi? |
| <50% | Stay local | "Fix this bug" → Claude handles it |
Explicit Delegation
Take control with the @cli syntax:
@kimi design a modal component
@gemini research best practices for authentication
@codex analyze this entire codebase for security issues
# Shorthand aliases
@k @g @c
Parallel Execution
Independent tasks run simultaneously:
You: @kimi create the UI components @gemini write the documentation
[kimi] Building Header.tsx... [gemini] Generating API docs...
[kimi] Building Sidebar.tsx... [gemini] Writing usage guide...
[kimi] ✓ Complete [gemini] ✓ Complete
Smart Fallbacks
If a CLI fails or isn't installed, MMCO gracefully falls back:
✗ codex → not found
Routing adjusted: codex tasks → claude fallback
Supported CLIs
| CLI | Strengths | Install |
|---|
| Claude | Coding, debugging, architecture | claude.ai/code |
| Kimi | Frontend, design, UI/UX, animations | uv tool install kimi-cli |
| Gemini | Research, search, documentation | npm i -g @anthropic/gemini-cli |
| Codex | Large context, codebase analysis | npm i -g @openai/codex |
Installation
# Add the marketplace and install the plugin
claude plugin marketplace add https://github.com/syedadeel2/multi-model-cli-orchestrator
claude plugin install multi-model-orchestrator@multi-model-cli-orchestrator
# Restart Claude Code to activate
Alternative: Manual installation
git clone https://github.com/syedadeel2/multi-model-cli-orchestrator.git \
~/.claude/plugins/multi-model-cli-orchestrator
Quick Start
-
Check status - See which CLIs are available:
/mmco-status
-
Delegate explicitly - Send a task to a specific CLI:
@kimi create a signup form with validation
-
Let it route - Just describe your task and MMCO picks the best tool:
Design a beautiful landing page with hero section
→ Auto-delegating to kimi (92% confidence)
Configuration
Global Config (~/.config/mmco/config.yaml)
general:
auto_delegate: true
parallel_execution: true
confidence_threshold:
auto: 80
suggest: 50
routing:
patterns:
- match: ["landing page", "hero section"]
cli: kimi
weight: 40
Project Override (.mmco.yaml)
routing:
default_cli: kimi # This project prefers Kimi for ambiguous tasks
overrides:
- match: ["api", "backend"]
cli: claude
How It Works
┌─────────────────────────────────────────────────────────┐
│ Claude Code (Host) │
│ ┌─────────────────────────────────────────────────────┐ │
│ │ MMCO Orchestrator Plugin │ │
│ │ ┌──────────┐ ┌──────────┐ ┌─────────────────┐ │ │
│ │ │ Router │ │ Registry │ │ Process Manager │ │ │
│ │ └────┬─────┘ └────┬─────┘ └────────┬────────┘ │ │
│ └───────┼─────────────┼─────────────────┼────────────┘ │
└──────────┼─────────────┼─────────────────┼──────────────┘
│ │ │
▼ ▼ ▼
┌──────────┐ ┌──────────┐ ┌──────────┐
│ Kimi │ │ Gemini │ │ Codex │
│ Frontend │ │ Research │ │ Analysis │
└──────────┘ └──────────┘ └──────────┘