Browser Use
Browser agents and stealth cloud browsers for AI: run web tasks, fill forms, extract data.
作者提供的说明(英文原文)
WCAG 2.1 AA accessibility scanner (Playwright + axe-core) with grounded, browser-verified fixes.
这三项永远不会合并成一个分数:意见、测量和人气是不同的东西。
我们还没有为它写通俗的介绍。下面是我们从注册中心读取到的事实。
选择你使用的 AI 应用,然后复制设置。
ALLCHEMY_API_KEY标为机密的值是你账户的钥匙。千万不要分享给别人,也不要粘贴到聊天里。
你需要
你用的是哪个应用?
在上方选择你的应用。我们会记住它,下次按钮会显示“安装到 …”。
打开 Claude Desktop → Settings → Developer → Edit Config,把下面的内容粘贴到 claude_desktop_config.json
{
"mcpServers": {
"aether-wcag-scanner": {
"command": "npx",
"args": [
"-y",
"@allchemylabs/aether-wcag-scanner"
],
"env": {
"ALLCHEMY_API_KEY": "your-value"
}
}
}
}在终端中运行这条命令
claude mcp add aether-wcag-scanner --env ALLCHEMY_API_KEY=your-value -- npx -y @allchemylabs/aether-wcag-scanner运行前请先填写空着的值。
粘贴到项目中的 .cursor/mcp.json,或 ~/.cursor/mcp.json
{
"mcpServers": {
"aether-wcag-scanner": {
"command": "npx",
"args": [
"-y",
"@allchemylabs/aether-wcag-scanner"
],
"env": {
"ALLCHEMY_API_KEY": "your-value"
}
}
}
}浏览器会先询问是否打开应用,然后应用会显示设置并请你确认。密钥的值会留空,由你自己填写。
粘贴到项目中的 .vscode/mcp.json
{
"servers": {
"aether-wcag-scanner": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"@allchemylabs/aether-wcag-scanner"
],
"env": {
"ALLCHEMY_API_KEY": "your-value"
}
}
}
}浏览器会先询问是否打开应用,然后应用会显示设置并请你确认。密钥的值会留空,由你自己填写。
这个服务器在你的电脑上运行,而 ChatGPT 只能连接互联网上的服务器。请选择其他应用。
你的选择通过 Cookie 保存在这台设备上,无需账户。
这些设置是根据作者在注册中心的条目生成的。如果 README 的写法不同,请以 README 为准。
它在你的电脑上运行;安装后,工具列表会显示在你的 AI 应用中。
我们检查的是任何人都能看到的信息;我们没有阅读代码。不适用的检查项不计入总数。 · 检查于 2026年11月7日
注册名称已通过 GitHub 账户 allchemylabs 验证,该账户拥有此仓库
我们还没有从 GitHub 读取这个仓库
我们还没有从 GitHub 读取这个仓库
它在你的电脑上运行,没有互联网端点
在你的电脑上运行,改用本地设置
它在你的电脑上运行;我们只能读取互联网端点的工具列表
与注册中心里这个分类中所有活跃的服务器比较。只有当发布者拥有该仓库时,星标数才计入。
它在 1 项适用的基础检查中通过了 1 项。我们没有阅读代码。在添加密钥或令牌之前,请先看看这个服务器要求哪些权限,并使用只具备所需权限的密钥。
你需要填写:ALLCHEMY_API_KEY
它在你的电脑上运行。你的 AI 应用会在需要时启动它,数据不会经过作者的服务器(除非它自己调用在线服务)。
GitHub 仓库最后更新于–,最新版本(1.0.8)于17天前进入注册中心。
支持 MCP 的应用,例如 Claude Desktop、Claude Code、Cursor 和 VS Code。你的电脑上还需要有 Node.js。
同一分类中有:Browser Use, Chrome DevTools MCP, MCP Server Browser
同一分类中其他作者的服务器,按星标数排序。
如何看卡片:★ GitHub 星标i三大支柱:评价 · 检查 · 使用情况i更新 = 最新版本发布到注册中心的时间i
Browser agents and stealth cloud browsers for AI: run web tasks, fill forms, extract data.
MCP server for Chrome DevTools
MCP server for browser use access
按原文从作者的仓库复制。PopNx MCP 没有审查或背书其内容。HTML、图片和脚本已被移除。链接会离开本站。
An MCP server for Claude Code, Cursor, Codex and other coding agents that finds WCAG 2.1 AA accessibility violations and returns ready-to-apply fixes — not just a list of problems.
It drives a real browser (Playwright + axe-core) across desktop, tablet, and mobile viewports, then pairs each violation with a concrete fix: corrected HTML, the relevant WCAG success criteria, and technique code examples.
It works free without a key. For higher quality, contextual fixes, get a beta key at [beta.allchemylabs.ai](https://beta.allchemylabs.ai/?src=github) (sign in with GitHub).
An AI coding agent working from markup alone has to infer what a page does. Aether runs the page in a real browser, measures it with axe-core, and grounds every suggested fix in the WCAG spec, so the agent is working from measurements rather than inference.
| Coding agent alone | Aether, offline | Aether, with API key | |
|---|---|---|---|
| How it looks at the page | Reads the HTML it is given | Renders the page in Chromium at desktop, tablet and mobile widths and runs axe-core | Same |
| What it can measure | Structure visible in the markup | What axe measures on the rendered page: contrast, accessible names, ARIA usage, focusable regions, landmarks | Same |
| Fix suggestions | From the model's general knowledge | Deterministic templates for the common rules | Templates first, then WCAG-corpus examples, then a model; every cited technique passes a grounding check |
| How fixes are checked | Not checked unless you set that up | axe re-run on the fixed element; reports what cleared and what regressed | Same |
| What each fix cites | Whatever the model recalls | The axe rule and its severity | The WCAG success criteria and techniques, with rejected candidates shown |
| Network | Depends on the agent | None; runs entirely on your machine | Fix requests go to the hosted engine; falls back to offline templates when it is unreachable |
/plugin marketplace add https://github.com/allchemylabs/claude-plugins.git
/plugin install aether-wcag-scanner@allchemylabs
Once the official Claude Code marketplace listing is live, you can also install it directly from there.
Installing with /plugin install inside Claude Code prompts for your Allchemy Labs API key; paste the key you got from beta.allchemylabs.ai. Installing from the terminal (claude plugin install …), or letting Claude run the install for you, does not prompt, so set the key afterwards with the command below. Run the same command to change the key later:
/plugin configure aether-wcag-scanner@allchemylabs
The first start installs the plugin's dependencies (about a minute) and the first scan downloads Chromium for Playwright. If /mcp shows the server as not connected right after install, wait a moment and run /reload-plugins.
The plugin asks for a key at install because the cloud fixes need one. The scanner itself runs locally and still works without a key (for example after a terminal install, or via npx): you get local template fixes for every violation. A beta key adds higher quality, contextual fixes from the hosted insights engine; get one at beta.allchemylabs.ai (sign in with GitHub). Set it via the prompt above, or put ALLCHEMY_API_KEY= in the .env of the project you are scanning.
The same server is published to npm and the MCP Registry as @allchemylabs/aether-wcag-scanner. Any MCP client that speaks stdio can run it:
{
"mcpServers": { "aether-wcag-scanner": { "command": "npx", "args": ["-y", "@allchemylabs/aether-wcag-scanner"], "env": { "ALLCHEMY_API_KEY": " " } } } }
The first scan downloads Playwright's Chromium once (about 500 MB on disk: the browser plus its headless shell). Without ALLCHEMY_API_KEY the scanner still runs fully offline and returns template fixes.
Claude Code gets the /wcag-scan skill, which drives the scan → review → fix → verify loop. Other agents only see the tool descriptions and the instructions the server sends at connect time, so for the same behaviour copy AGENTS.md into your project (Codex reads it natively; Cursor and Windsurf accept it as a rules file). Without it an agent may fetch the page HTML and guess instead of running the browser scan.
| Tool | What it does |
|---|---|
aether_scan_and_fix | Recommended. Scan a URL — or a single-page app across routes via the spa block — and return violations *with* fixes, each verified on the real browser, in one call. Set maxFixes: 0 for a fast scan-only pass. |
aether_get_fix | Get a fix for one specific violation (verified on the real browser). Omit html to instead get a rule-level explanation — WCAG success criteria and technique examples. |
aether_verify_fix | Deterministically verify a fix with real Playwright + axe — re-scan a URL or diff original vs fixed HTML. Reports whether the target cleared, any new regressions, and the net compliance delta. |
aether_check_html | Analyze an HTML snippet for issues — no live URL or browser needed. |
aether_submit_feedback | Rate a fix (useful / not useful) so the guidance engine keeps improving; your correction helps seed better fixes. |
Just ask Claude naturally, e.g. *"check localhost:3000 for accessibility issues and fix them"* — it will pick the right tool.
/wcag-scan skillRun /wcag-scan for a guided workflow: scan → review violations → apply fixes → re-scan to verify → summarize. The skill orchestrates the tools above and edits your files in place.
The plugin adds two hooks to Claude Code:
To turn the post-edit check off in every project, run /plugin configure aether-wcag-scanner@allchemylabs and set Post-edit accessibility check to false. To turn it off (or back on) for one project, add this to that project's .claude/settings.local.json:
{ "env": { "AETHER_POSTEDIT_HOOK": "0" } }
Use "1" to turn it back on for that project. Restart Claude Code after either change.
Pass a spa block to aether_scan_and_fix instead of url:
{ "spa": { "entryUrl": "http://localhost:4200", "framework": "angular", "maxRoutes": 10 } }
entryUrl — where the app boots. Routes are discovered by crawling client-side links unless you list them in routes: [{ "path": "/settings", "name": "Settings" }].framework — angular, react, vue, or omit to auto-detect.maxRoutes — cap on discovered routes.Navigation is client-side (the browser context and app state are kept between routes), and readiness is decided by a layered stability cascade — DOM quiet, framework hooks such as Angular testability, pending-request drain — never by Playwright's networkidle, which deadlocks on apps with long-polling or websockets. A config that sets networkIdle is rejected for that reason.
Aether is an accessibility analysis and guidance tool. Using the platform does not, by itself, make your website, application, or digital service conformant with WCAG, the ADA (including Title II), the European Accessibility Act, Section 508, EN 301 549, or any other accessibility law or standard.
Automated scans detect only a subset of accessibility barriers. A "zero issues" result means the scanner did not flag anything in its detection range — not that the property is legally compliant.
You remain responsible for:
Aether does not provide legal advice and does not replace an accessibility audit or a VPAT/ACR prepared by a qualified auditor.
MIT for the client (this plugin). The hosted insights engine and related Allchemy Labs services are proprietary — see LICENSE and NOTICE.