Myagi - Open Agent Registry

bot.myagi/openagent-registry

Source

Documentation

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):

Official MCP Registry - the MCP working group's catalog of MCP servers (+25 if your domain ships one)
Standards: llms.txt, agents.txt / agents.json, A2A Protocol, Web Bot Auth, RSL

Community resources used for domain discovery only (no score credit; we use these to know which sites to scan):

thedaviddias/llms-txt-hub - one maintainer's community list of sites publishing llms.txt
Agent Friendly Directory - sklivvz's curated agent-usable services
A2A Registry - community directory of A2A agents
LLMTEXT - llms.txt validation toolkit

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

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.