ARA (Agent-Ready Architecture) is an open web standard that lets any website declare itself understandable and actionable by autonomous AI agents — in a single HTTP request.
{
"$ara": "1.0",
"identity": {
"name": "Your Site",
"type": "ecommerce",
"description": "Online store with 2,000 products"
},
"content_map": { /* resources, schemas, endpoints */ },
"capabilities": { /* MCP, A2A, REST, GraphQL */ },
"policies": { /* rate limits, auth, data usage */ }
}Today, AI agents waste 90% of their tokens just trying to understand a website.
ARA reduces token consumption by 10-25x
Existing standards serve crawlers and search engines. None of them serve autonomous agents that need to understand, decide, and act.
Binary allow/disallow. No semantics. Tells bots where they can't go, not what they can do.
A list of URLs. No context about what each page contains or how to interact with it.
Fragmented per page. No global view. An agent must crawl the entire site to build understanding.
Plain text, unstructured. No schemas, no actions, no protocol negotiation. A step in the right direction, but not enough.
Technical API docs for developers. No business context, no intent mapping, no agent policies.
Complete: identity, schemas, actions, protocols, policies. An agent understands any site in one request.
ARA gives every website a structured, machine-readable interface for AI agents.
What can agents do? Actions with typed inputs/outputs, intent mapping, workflow sequencing.
actions.json
How is data structured? Semantic schemas with search hints and cross-resource relationships.
schemas/*.json
Who is this site? What does it offer? How to connect? One file, one request, full understanding.
manifest.json
Agents choose the best communication channel based on their capabilities.
| Capability | robots.txt | sitemap | Schema.org | llms.txt | OpenAPI | ARA |
|---|---|---|---|---|---|---|
| Site discovery | — | ~ | — | ~ | — | ✓ |
| Global overview | — | — | — | ~ | — | ✓ |
| Data schemas | — | — | ~ | — | ✓ | ✓ |
| Actions / interactions | — | — | — | — | ✓ | ✓ |
| Intent mapping | — | — | — | — | — | ✓ |
| MCP / A2A support | — | — | — | — | — | ✓ |
| LLM-optimized digest | — | — | — | ~ | — | ✓ |
| Agent policies | ~ | — | — | — | ~ | ✓ |
✓ = Full support ~ = Partial — = Not supported
Real-world manifest examples for different site types.
You don't need to implement everything. Even Level 1 transforms agent interactions.
Agent parses raw HTML. Slow, fragile, expensive. ~20,000 tokens per page.
Full discovery + global understanding. Token cost drops to ~800. 90% improvement.
Agents understand your data structure. Cross-site comparison becomes possible.
Agents can search, book, order. Programmatic interaction without UI scraping.
Native agent experience. Your site becomes a first-class citizen in the agentic web.
Validate, generate, and adopt ARA with command-line tools.
Check if a site is ARA-ready and get a score.
$ npx ara-validate https://mysite.com
✓ manifest.json found
✓ identity — valid
✓ content_map — 4 resources
⚠ actions.json — 1 action missing output
✓ digest.md — 287 tokens
Score: 92/100 — Grade: AAuto-generate a manifest from any existing site.
$ npx ara-generate https://mysite.com
✓ Fetched main page (42KB)
✓ Extracted metadata: "MyShop"
✓ Detected type: ecommerce
✓ Found 3 JSON-LD blocks
✓ sitemap.xml found
✓ Generated ARA manifest (Level 1)Three steps. Five minutes. One file.
mkdir -p .well-known/araStart from an example or generate one automatically:
npx ara-generate https://yoursite.com > .well-known/ara/manifest.jsonnpx ara-validate https://yoursite.comThat's it. Your site is now discoverable by AI agents.