# Ora

> Scores how well your site works for an agent, from the terminal or in CI

Category: MCP servers (https://aiarabicdev.com/mcp-servers)
Canonical page: https://aiarabicdev.com/mcp-servers/ora
Tags: mcp, seo, audit, testing, agents

Ora fetches a site the way an agent would and grades what it could find, read and actually use. The scoring service is hosted and closed, but the ax CLI that drives it is MIT and needs no account, so an audit is one npx command and a build can fail on the number.

## Facts

- Licence: MIT for the ax CLI, the scanning service itself is closed
- Language: TypeScript
- Cost: Free core, paid tiers exist
- Self hostable: no
- Added: 2026-09-08
- Last checked: 2026-09-08

## Links

- Website: https://ora.ai
- Source: https://github.com/ora/ax
- Docs: https://ora.ai/docs

## What the score is actually made of

Four weighted layers make up the hundred points: discovery at twenty, access at thirty, usability at forty and payments at ten. Usability carries the most because an agent that found you and read your documentation still stops dead at a login flow or an endpoint nobody wrote down.

Payments is the odd pillar. Ten points ride on accepting machine payments over rails like x402, which very little of the web does today. Checks carry an applicability flag, so run the check catalogue before you treat a missing ten as work you owe.

## The two ways in

From a terminal, npx ax audit prints the score and every failing check with a fix. Add a minimum score and it exits non zero, which is the whole CI story. A tunnel command lets it reach localhost, so you can measure a branch before it is public rather than after.

As an MCP server it is a streamable HTTP endpoint with no key for reads, and your own agent gets the interesting verbs: scan a domain, re run only the checks you just fixed, read the leaderboard, fetch a published skill. The deep journey command is the part worth watching, because a recording of an agent giving up on your checkout tells you more than a grade does.

## What you hand over to use it

The scan runs on their infrastructure, not yours, so there is no version of this you host and no way to read why a check decided what it decided. You are taking one vendor's opinion of agent readiness and its weights, which is fine as a direction of travel and thin as an audit.

The leaderboard is public and ranks tens of thousands of domains by grade. The documentation only promises that tunnel and ephemeral scans stay off it, so assume scanning a real domain can attach a public number to it, including a domain that is not yours.

## Reach for it when

- you ship something you expect agents to read, integrate with or buy from
- you want a number you can gate a build on rather than another written audit
- you would rather watch an agent fail on your site than guess where it gets stuck
- you already publish llms.txt and want to know whether it changed anything

## Watch out for

- only the ax CLI is open source, the scanner is hosted and closed, so you cannot run the scoring yourself or inspect how a check reached its verdict
- the leaderboard is public and the docs exempt only tunnel and ephemeral scans, so scanning a real domain can put a grade on it in public
- keyless scanning is capped at thirty scans per rolling twenty four hours per IP and six forced rescans, and scan keys are issued by hand after you contact them, so unattended CI is not a given
- it is very young, the CLI is still on 0.x and the repository is weeks old, so expect flags and output shapes to move under you

## Get started

Score a site:

```sh
npx ax audit https://your-site.com
```

Gate a build on the score:

```sh
npx ax audit https://your-site.com --min-score 70
```

Add the MCP server to Claude Code:

```sh
claude mcp add --transport http ora https://ora.ai/api/mcp
```

---

aiarabicdev is a free directory of open source tools. No affiliate links, no sponsored placements.
New entries are announced at https://t.me/arabic_Intelligence