impeccable-mcp
MCP server that wraps impeccable's design-lint engine to scan URLs and raw HTML for 59 deterministic anti-patterns, returning structured findings grouped by slop vs quality via headless Chrome.
README
<img src="assets/fathom-header-banner.svg" alt="Fathom Works — impeccable-mcp" width="100%">
$ impeccable-mcp
An MCP server wrapping impeccable's design-lint engine — headless Chrome plus 59 deterministic anti-pattern rules — so any agent can scan a URL and get back structured findings, no per-agent setup required. No LLM, no API key needed for the scan itself; the engine is pure detection.
[ what it does ]
- Takes
impeccable's public library exports (detectUrl,detectHtml,ANTIPATTERNS,createBrowserDetector) — an official npm dependency, not a fork — and exposes them as MCP tools - Runs one long-lived headless Chromium instance, each scan in its own incognito context
- Serves MCP over streamable HTTP on port 8000, same shape as gsd-browser-mcp
- Lets any MCP client (Claude Code, Fred, Billy, whatever's next) scan a website's design without local setup
[ tools ]
| Tool | Needs a browser | What it does |
|---|---|---|
scan_url |
yes | Loads a URL in headless Chrome, runs all detector rules, returns findings grouped by slop (AI design tells) vs quality (a11y/design issues). Accepts optional headers / cookies for authenticated pages. |
scan_html |
no | Runs the static detector against a raw HTML/CSS string. No browser launched — use this to check generated markup before it's written to disk. |
screenshot |
yes | Returns a PNG of the rendered page as MCP image content. |
list_rules |
no | Returns the full rule registry (id, name, category, description) from impeccable's ANTIPATTERNS. |
Header/cookie values you pass to
scan_urlwill appear in the calling agent's transcript. That's a deliberate tradeoff for reaching authenticated pages, not an oversight — seeCONTRIBUTING.md. Don't pass anything you wouldn't want logged.
The private-target guard
scan_url and screenshot refuse RFC1918, loopback, and link-local targets by default. This
service's whole job is "fetch whatever URL an agent hands it" — without the guard, a
prompt-injected agent could turn it into a LAN port scanner or point it at Portainer/Home
Assistant from inside the trusted network. Set IMPECCABLE_ALLOW_PRIVATE=1 only if you specifically
need to scan an internal site and understand the tradeoff.
[ quick start ]
1. pull and run
$ docker run -d \
-p 8000:8000 \
--shm-size=512mb \
-e IMPECCABLE_MCP_TOKEN=$(openssl rand -hex 32) \
ghcr.io/jemplayer82/impeccable-mcp:latest
2. connect your mcp client
http://localhost:8000/mcp
with Authorization: Bearer <IMPECCABLE_MCP_TOKEN>.
[ docker compose ]
services:
impeccable-mcp:
image: ghcr.io/jemplayer82/impeccable-mcp:latest
restart: unless-stopped
shm_size: "512mb"
mem_limit: 1536m
init: true
ports:
- "8000:8000"
environment:
IMPECCABLE_MCP_TOKEN: your-token-here # pragma: allowlist secret — doc placeholder
healthcheck:
test: ["CMD", "curl", "-sf", "http://localhost:8000/healthz"]
interval: 30s
timeout: 5s
retries: 3
start_period: 20s
[ environment variables ]
| Variable | Required | Description |
|---|---|---|
IMPECCABLE_MCP_TOKEN |
Yes | Bearer token clients must send. Server refuses to start if unset. |
PORT |
No | Listen port inside the container. Default 8000. |
IMPECCABLE_ALLOW_PRIVATE |
No | Set to 1 to allow scans of RFC1918/loopback/link-local targets. Default off. |
MAX_CONCURRENCY |
No | Max concurrent browser-backed scans. Default 2. |
BROWSER_IDLE_TIMEOUT_MS |
No | Close the idle browser after this long with no scans. Default 300000 (5 min). |
[ how it works ]
A single Node/ESM process:
@modelcontextprotocol/sdk'sStreamableHTTPServerTransportover rawnode:http— stateless, a fresh server + transport per request, no session bookkeeping.src/scan.mjsholds one Chromium viaimpeccable'screateBrowserDetector, relaunching on crash and closing afterBROWSER_IDLE_TIMEOUT_MSof inactivity. Each scan gets its own incognito browser context so headers/cookies from one call never leak into another.- Findings come back in
impeccable's stable shape (antipattern,name,description,severity,category,snippet), grouped by category with a count summary.
impeccable itself needs no LLM and no API key — it's a deterministic rule engine. This server
doesn't add one either; it's transport, auth, browser lifecycle, and a URL guard around a library.
[ related ]
- impeccable — the underlying detection engine (don't
file issues/PRs there from this project; see
CONTRIBUTING.md) - gsd-browser-mcp — the sibling MCP this repo's transport, auth, and Dockerfile pattern were cloned from
- Impeccable Chrome extension — same rule engine, for a human looking at DevTools directly. This server exists for agents; the extension is still the right tool for eyeballing a page yourself.
[ license ]
Apache License 2.0 — see LICENSE.
<img src="assets/fathom-footer-banner.svg" alt="Fathom Works — sound the depths before you set a course" width="100%">
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。