Getting Started

This guide gets Radar running from the CLI first. After that, connect it to your AI agent through MCP or project rules.

Prerequisites

• Node.js 20 or newer.
• pnpm, npm, or another package manager that can install npm packages.
• An API key for an OpenAI-compatible worker model provider.

Install Radar

pnpm install -g @contextrail/radar

Configure A Worker Model

Set an API key:

export RADAR_API_KEY=sk-...

Radar defaults to Moonshot / Kimi:

export RADAR_BASE_URL=https://api.moonshot.ai/v1
export RADAR_MODEL=kimi-k2.5

Any OpenAI-compatible chat completions endpoint can work.

Try The CLI

Ask Radar to read a couple of files and return a short answer:

radar ask -p package.json README.md -q "What is this package for?"

Summarize a long test output:

pnpm test 2>&1 | radar summarize --stdin -f "focus on failures and root causes"

Draft a file without writing it:

radar write --dry-run \
  -s "test file matching this style" \
  -c test/service.test.ts \
  -t test/new-feature.test.ts

Add Agent Routing Rules

Installing Radar does not automatically modify your project. Copy the routing rules you want into the consuming project:

• AGENTS.md for general agent guidance.
• .cursor/rules/radar-delegation.mdc for Cursor.
• CLAUDE.md for Claude Code.

See MCP Clients and the integration guides for client-specific setup.

Next Steps

• Read the Delegation Model.
• Add an MCP client config from MCP Clients.
• Try the Documentation Updates recipe.