Try it

# Point it at a site. See what fails, and how sure Marlo is.

A web address, a GitHub repository, or a paste of HTML. Marlo's own rules run here, in your browser, and every finding carries the rule's measured precision.

any public site works. go on. 

 Check it

A web address is fetched once by this site, which hands back that page's HTML and keeps nothing. If the site refuses automated requests, or builds itself with JavaScript, a real browser in Cloudflare loads it instead. A repository is read straight from GitHub, up to twenty-five of its HTML files. The preview never runs the page's scripts.

Try: Deque's deliberately inaccessible demo The W3C's inaccessible demo GOV.UK, a well-built real site Marlo's demo pages on GitHub

Loading Marlo's rules.

The playground runs Marlo in your browser, which needs JavaScript. Everything else on this site works without it.

Or paste HTML 

Pasted HTML never leaves this page. Relative links in it go nowhere.

Check the paste

Examples: Broken checkout Broken settings A clean page

Preview, with the page's scripts off

Download the report as JSON

What runs here

## One engine of four, and it says so

This is Marlo's own rule engine, the same code `marlo scan --renderer browser` injects into Chromium. The command line also runs axe-core, Alfa and HTML CodeSniffer, routes each rule to the engine measured most precise, and refuses to report a rule clean when any engine failed it. None of that happens here, so this page finds less than the command line does, and its misses are Marlo's own. The [numbers page](https://trymarlo.pages.dev/accuracy/) says how often Marlo's engine is wrong, on the official cases and on real pages.

A rule that needs a person to decide says so instead of guessing, and a rule that could not run is listed as not examined. Neither is a pass.

Any page you open

## The bookmarklet

works on pages built with JavaScript, too Drag this to your bookmarks bar, open any page, and click it: [Check with Marlo](javascript:(()=>{const s=document.createElement('script');s.src='https://trymarlo.pages.dev/try/marlo-page.js';s.onerror=()=>alert('Marlo could not load here. This page may forbid scripts from other sites. You can paste its HTML into https://trymarlo.pages.dev/try/ instead.');document.documentElement.appendChild(s)})())

It loads Marlo's rules from this site, checks the page as your browser has rendered it, and opens a panel listing each failure. Choose an element in the panel to outline it on the page. It sends nothing back. Some sites forbid scripts from other origins, and on those it will say it cannot run, which is the site's security policy doing its job.

Beyond the browser

## Every other way in

### Command line

All four engines, routing, the invariant and SARIF. No network after install.

```
git clone https://github.com/KarthikSubramanian07/Marlo.git
cd Marlo && pnpm install && pnpm build
node packages/cli/dist/bin.js scan page.html
```

### Docker

The command line with nothing to install but Docker. Mount the folder to scan.

```
docker run --rm -v "$PWD:/work" ghcr.io/karthiksubramanian07/marlo scan page.html
```

### GitHub Action

Read-only by default. Results land in the code scanning tab as SARIF.

```
- uses: KarthikSubramanian07/Marlo/.github/actions/marlo@main
```

[Inputs and the opt-in pull request mode](https://github.com/KarthikSubramanian07/Marlo/tree/main/.github/actions/marlo)

### MCP, for coding agents

A remote, read-only endpoint for coverage and per-rule accuracy at `https://trymarlo.pages.dev/mcp`, and a local server, `@trymarlo/mcp`, that scans the markup an agent just wrote. Neither has a tool that writes.

[Developer portal](https://trymarlo.pages.dev/developers/)
