website-audit
Enables auditing any public website, returning a scored plain-English report that flags issues costing customers, covering speed, phone experience, search visibility, contact options, writing quality, and modernity.
README
Website Audit MCP
An MCP server that audits any public website and reports, in plain English, what is costing its owner customers.
Ask your assistant to audit a site and you get a scored report back:
audit therecoveryoasis.com
Website audit: https://www.therecoveryoasis.com/
Score: 50/100 — 4 things on this page are costing you customers.
Speed — 0%
✗ The HTML alone is 920 KB
Before a single image or script, the page document itself is this big.
✗ One image on this page is 3.2 MB
Oversized images are the most common reason a site feels slow on a phone.
Getting in touch — 0%
✗ Nobody can tap to call you
There's no tap-to-call link on this page. On a phone, every extra step
between someone and your number costs you calls.
How the writing reads — 58%
! The writing leans generic
A few stock marketing phrases showed up: holistic approach, transformative.
Install
Requires Node 18 or newer.
Claude Code
claude mcp add website-audit -- npx -y github:rolledoatmeals/website-audit-mcp
Claude Desktop, Cursor, or anything else that speaks MCP — add this to the config file:
{
"mcpServers": {
"website-audit": {
"command": "npx",
"args": ["-y", "github:rolledoatmeals/website-audit-mcp"]
}
}
}
Or clone it and run locally
git clone https://github.com/rolledoatmeals/website-audit-mcp
cd website-audit-mcp && npm install
claude mcp add website-audit -- node "$(pwd)/index.js"
Then ask: "audit example.com".
It works out what kind of site it is
A missing phone number is a serious finding for a body shop and a non-issue for a portfolio. So the tool works out whether it is looking at a local business, an online store, a portfolio, or a software product, and scores accordingly. It tells you what it decided and why:
Type: Local business — detected from AutoRepair schema markup
If it guesses wrong, pass the type yourself:
audit example.com as a portfolio
Online stores get product-schema and returns-policy checks. Software products get checked for a clear sign-up path and visible pricing. Portfolios get checked for shown work and profile links, and are not penalised for having no street address.
What it checks
Six categories, each weighted so one weak area cannot sink the whole score.
| Category | Looks at |
|---|---|
| Speed | HTML size, total download weight including scripts and stylesheets, oversized images, compression, server response time |
| Phone experience | Viewport, whether the page is blank until JavaScript runs, missing alt text |
| Getting found | Title and description quality, H1, structured data, Open Graph tags, sitemap |
| Getting in touch | Tap-to-call, contact form, email address, a readable street address, HTTPS |
| How the writing reads | Stock AI-marketing vocabulary, em-dash density, "not just X but Y" patterns, exclamation spam, leftover placeholder text |
| How current it looks | <center> and <font> tags, table layouts, Flash remnants, ancient jQuery, Bootstrap 3, modern image formats, lazy loading |
It also detects the platform a site is built on, matching asset URLs rather than page copy so a site that merely mentions a platform by name is not misread.
Notes on accuracy
JavaScript-rendered pages are detected and flagged. A single-page app serves an almost empty HTML document, which would otherwise score well for being small. When that is detected the tool says so and skips the checks it cannot answer honestly, rather than reporting a flattering number.
Weight is measured, not guessed. HTML, scripts, stylesheets, and images are each fetched to read their real size.
One page at a time. This audits the URL you give it, not the whole site.
Security
The server fetches URLs you name, so it refuses anything that could reach a private network:
- HTTP and HTTPS only, ports 80 and 443 only
- DNS is resolved and private, loopback, link-local, and reserved ranges are rejected
- Every redirect hop is revalidated, so a public URL cannot redirect into a private one
- Response bodies are capped and all requests time out
Built by
Zachary Shepelsky — AI operations and automation, Tampa FL.
The same engine runs as a web tool at zacharyshep.com/audit.
MIT licensed.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.