Run it
npx scandiary https://your-site.com npx scandiary localhost:3000
It opens your pages in the Chrome, Edge or Brave already on your machine, runs
axe-core against WCAG 2.0 and 2.1 at levels A and AA, and lists what it found,
page by page. Five pages by default, up to a hundred with --pages.
It needs Node 20 or newer.
Get a fix prompt
npx scandiary https://your-site.com --prompt > fix-a11y.md
Paste that file into your coding tool. Each issue comes with real examples from your pages and a note on how to fix it in the source. The prompt also says what not to do: no overlay scripts, no hiding elements just to pass the check, list any label it had to guess, and do not call anything done until a new scan agrees.
Text copied from your pages goes into the prompt as data, not as instructions. It is flattened, shortened and marked so your tool should not follow anything written inside it. Read what your tool changes, as you would with any generated code.
Inside Claude Code
Install the skill once:
mkdir -p ~/.claude/skills/scandiary curl -o ~/.claude/skills/scandiary/SKILL.md https://raw.githubusercontent.com/Yedigarli/scandiary-cli/master/skills/scandiary/SKILL.md
Then ask in plain words, for example “check the accessibility of localhost:3000 and fix what you find”. Claude runs the check, edits your code, runs the check again and reports the numbers before and after.
As an MCP server, for Cursor and others
Add this to your client’s MCP settings:
{
"mcpServers": {
"scandiary": {
"command": "npx",
"args": ["-y", "scandiary", "--mcp"]
}
}
}
It adds one tool, check_accessibility, which takes a URL and returns
the fix prompt. In Claude Code the same thing is one line:
claude mcp add scandiary -- npx -y scandiary --mcp. It is listed in
the official MCP registry as io.github.Yedigarli/scandiary.
We have tested it in Claude Code and with the official MCP client library. The settings above are the standard MCP format, but we have not yet tried them in Cursor or Windsurf ourselves.
What we measured
On 3 November 2026 we gave a coding agent our own demo shopping cart page, with three icon buttons that had no name and a photo with no text alternative. It had only the page and the request to check and fix it.
4
elements found by the first scan: three unnamed buttons, one image
0
found by our own second scan, after the agent edited the page
We ran it three ways: pasting the prompt, the Claude Code skill, and the MCP tool alone. Each came back with the same count. In each run the agent added names and text alternatives and added no script. In the first two it listed the labels it had guessed. In the MCP run it did not, so that rule is now written into the prompt itself.
One small page, one run per way. That shows the prompt works on this kind of problem. It does not show how it does on a large app, or with other tools.
What it cannot tell you
Automated checks find roughly a third of accessibility problems, this one included. The rest needs a person with a keyboard and a screen reader.
A clean result means these rules found nothing on the pages checked. It does not mean your site is accessible, and it does not say whether you meet any legal standard.
The same check, every Monday
A one-off scan tells you where you stand today. A deploy next month can break it again without anyone noticing. Scandiary runs the check every week and emails you what changed, so you keep a dated history. Twenty-five pages are free.