Ninja CLI Documentation

Terminal-native multi-agent coder. Plan with a model you trust, run work with local or cloud workers, keep approvals under your control. This site covers install, model setup, CLI usage, MCP, and tool API keys.

Quick links

Installation

One-line install for Linux, macOS, and Windows.

Quickstart

Point Ninja at a model and start coding.

Configure models

Planner, coder, reviewer, and hybrid routing.

config.json

Where settings live and how precedence works.

CLI commands

session, worktree, policy, doctor.

Flags

Headless mode, --yes, and common switches.

MCP

Connect Model Context Protocol servers.

Tools & API keys

Serper search keys, enabled_tools, MCP secrets.

What is Ninja?

Ninja is a multi-agent coding CLI:

A planner reads your project and coordinates work.
Subagents (coder, reviewer, and others) handle focused tasks.
Approvals and policy keep shell and tools under your control.
Optional MCP servers and skills extend what the agent can do.

It works with any OpenAI-compatible chat API (local servers, LiteLLM, major cloud providers, …).

Install in one line

# Linux / macOS
curl -sfL https://ninja-cli.btaskee.work/install.sh | bash

# Windows (PowerShell)
irm https://ninja-cli.btaskee.work/install.ps1 | iex

Then:

export OPENAI_BASE_URL="http://localhost:8080"
export OPENAI_MODEL="your-model"
cd your-project && ninja

Docs map

Section	Topics
Getting started	Install, first run
Configuration	Models, env vars, `config.json`
CLI	Commands and flags
Integrations	MCP, tool keys, web search
Safety	Approvals and headless policy
Help	Connection troubleshooting

Continue with Installation → (includes uninstall / remove).