OpenAgent Registry
The open registry, scanner, and leaderboard for agent-ready websites and services.
OpenAgent Registry (OAR) does three things in one place:
- Scans any URL for the full stack of agent-access signals -
llms.txt, agents.txt / agents.json, AGENTS.md, .well-known/agent-card.json, robots.txt + Content Signals, OpenAPI, OAuth discovery, RSL licensing, MCP/A2A registry presence - and aggregates upstream registries instead of re-doing their work.
- Scores each site against a transparent, public rubric and ranks it within its industry category.
- Tells you what to do about it - a concrete approval-path checklist with deep links to the actual developer applications, OAuth signups, partner programs, and standards-adoption steps. Lead with the checklist; the score is supporting evidence.
Sources of authoritative scoring credit (registries / standards bodies):
Community resources used for domain discovery only (no score credit; we use these to know which sites to scan):
What OAR adds that none of the above does today:
- Comparative leaderboards by industry category (rentals, shopping, content, social, developer)
- Approval-path generator - concrete next steps, not just "you have OAuth"
- Aggregation across the full stack in one report
- Change monitoring over time
- A public, free API so agents (including your own) can query it before calling external services
Status
Pre-alpha but live. Public deploy: https://openagent-registry.vercel.app.
After the v0.1 import, the registry tracks 1,400+ services across 6 industry categories, sourced from the original seed plus an import of llms-txt-hub. Every service has a live score, a label, and a generated approval-path checklist. Looking for contributors - especially for approval-path data per service (the moat).
Stack
- Next.js 15 (App Router) + TypeScript + Tailwind CSS
- No backend service in v0 - Next.js API routes do the scanning
- Local JSON cache for MVP; Postgres or Vercel KV for production scale
- Deploys natively to Vercel
Quickstart
pnpm install
pnpm dev
Then open http://localhost:3000.
How scoring works
See docs/rubric.md. Every signal is weighted, every weight is published, every score is reproducible from the JSON profile. No black-box scoring.
The current rubric (summary):
| Signal | Points |
|---|
| Official MCP server in MCP Registry | +25 |
| A2A Agent Card found | +25 |
agents.json (root or .well-known) | +15 |
| OpenAPI spec found | +15 |
| Web Bot Auth / signed-agent support | +15 |
| Official developer docs | +10 |
| Partner / approval path documented | +10 |
| OAuth/OIDC discovery | +10 |
agents.txt standard | +10 |
llms.txt | +8 |
AGENTS.md | +8 |
| RSL / licensing manifest | +8 |
robots.txt allows docs/API | +5 |
| Cloudflare Content Signals present | +5 |
llms-full.txt (extended) | +5 |
| Listed in llms-txt-hub | +5 |
| Listed in Agent Friendly Directory | +5 |
ai-plugin.json (deprecated signal) | +2 |
| Explicit bot/agent prohibition | -50 |
| Blocks automation, no API alternative | -30 |
| Unknown terms / can't parse | -10 |
Labels:
| Score | Label |
|---|
| 80–100 | Agent-ready |
| 60–79 | Agent-friendly |
| 40–59 | Partially agent-accessible |
| 20–39 | API-only / limited |
| 0–19 | Unknown |
| <0 | Likely blocked |
Categories
- Rentals & Travel - Airbnb, Booking, Hertz, Avis, Enterprise, Expedia, Vrbo, AutoRentals
- Shopping & Commerce - Amazon, Shopify, Stripe, Target, Walmart, BestBuy, Etsy
- Content & News - NYT, WaPo, Bloomberg, Reuters, Substack, Medium
- Social & Community - Reddit, X, LinkedIn, Bluesky, Mastodon, Discord
- Developer Platforms - GitHub, Vercel, Cloudflare, npm, AWS, plus 800+ dev tools from the upstream import
- AI & ML Platforms - Anthropic, OpenAI, and 500+ AI/ML platforms publishing
llms.txt
Within each category, services are ranked by their AgentRank score. The category page is the SEO-friendly editorial surface: "Top agent-ready rental sites, [month] [year]".
Contributing
See CONTRIBUTING.md. The three highest-leverage contributions today:
- Add services to
data/seed.json (a category, a homepage URL, optional notes) - opens a PR; CI runs a scan against it.
- Tighten parsers for the well-known files. The MVP parsers are intentionally conservative - better parsers raise the data quality for everyone.
- Approval-path data - concrete next-step instructions per service (developer signup URL, OAuth scopes needed, partner-program form). This is the killer feature; quality is the moat.
License
Apache 2.0.
