ARA (pronounced: ah-rah) is a package registry and distribution system for AI development artifacts. It provides a standardized way to discover, share, and install reusable components that enhance AI assistants and development workflows.
ARA is a vendor-agnostic open standard — any AI coding tool can implement it, including Kiro, Claude Code, Cursor, VS Code Copilot, Windsurf, and others.
ARA supports the following package types:
| Type | Description |
|---|---|
agent |
AI assistant configurations with prompts, tools, and behaviors |
mcp-server |
Model Context Protocol servers that extend AI capabilities |
powers |
Bundles of MCP tools, steering files, and hooks |
steering |
Persistent project knowledge files that guide agent behavior |
skill |
Procedural knowledge via SKILL.md |
context |
Knowledge files, prompt templates, and reference materials |
agents-md |
AGENTS.md format for guiding coding agents |
Use the optional platform field to declare tool-specific installation targets (e.g., "platform": "kiro").
- Vision - Understand the project goals and tenets
- ara.json Spec - Package manifest format
- JSON Schema - Validate your ara.json
- Registry API - HTTP API specification for registry implementations
- Security - Security guidance for implementers
{
"$schema": "https://raw.githubusercontent.com/ara-registry/spec/refs/heads/main/ara.schema.json",
"specVersion": "1.0",
"name": "acme/code-reviewer",
"version": "1.0.0",
"description": "AI agent for automated code review",
"author": "[email protected]",
"tags": ["code-review", "automation"],
"type": "agent"
}See CONTRIBUTING for more information.
This project is licensed under the Apache-2.0 License.