What is the Lighthouse MCP server?
The Lighthouse MCP server by danielsogl wraps Google Lighthouse so an AI agent can audit a page for performance, accessibility, SEO and security. It reports Core Web Vitals, finds unused JavaScript and checks performance budgets. You run it with npx @danielsogl/lighthouse-mcp@latest on Node 22 or newer.
It is a Model Context Protocol server that runs Google Lighthouse for an AI agent. The agent gives it a URL and gets back scores and recommendations across performance, accessibility, SEO and security, plus resource and Core Web Vitals analysis.
Every tool declares an output schema and returns validated structured content, so the client receives typed data rather than a JSON string to parse. Chrome or Chromium is needed, and the README says Lighthouse manages it automatically.
What it changes in your workflow
- The agent measures a page before and after a change, so a speed fix is backed by a Lighthouse number.
- Core Web Vitals tools report LCP, INP and CLS and can check them against thresholds you set.
- A budget tool checks a page against a performance budget you supply.
- Resource tools find unused JavaScript and analyze scripts, styles, images and fonts for savings.
- Lighthouse 13 audits cover how well a page serves AI agents, including WebMCP tools and llms.txt.
Local setup at a glance
| Lighthouse MCP server | |
|---|---|
| Where it runs | On your machine, started by the client with npx |
| Transport | stdio |
| Requirements | Node.js 22.0.0 or newer and Chrome or Chromium |
| Login sessions | Persistent Chrome profile with --profile-path and --no-headless |
| Account or key | None |
Who it is for
Best for
- Developers and site owners who want an agent to audit a page, then fix what Lighthouse flags.
- SEO and performance reviews that need Core Web Vitals and SEO checks in one tool.
- Pages behind a login, using a persistent Chrome profile run in headed mode.
Not for
- Real-user field data. This runs Lighthouse lab audits against a URL you give it.
- Fixing issues on its own. It reports findings and recommendations; your agent or you still make the changes.
- Deep accessibility sign-off. For a dedicated WCAG workflow see the A11y Ally skill.
Use it with UiChemy
After you build a page with UiChemy, point the server at its URL and ask for a performance and SEO audit. Composer writes HTML, CSS and JS for Elementor or Gutenberg; Figma to WordPress converts designs. Click Get Started on uichemy.com. For fixes, see the web performance skill.
Connect Lighthouse MCP Server: Web Audits for AI Agents to your AI client
npx @danielsogl/lighthouse-mcp@latestCheck requirements
You need Node.js 22.0.0 or newer and Chrome or Chromium installed.
Add the config
Put the config above in your client's MCP settings. The README gives one-click or manual steps for VS Code, Cursor, Windsurf and Claude Desktop.
Optional: keep a login
To audit signed-in pages, add --profile-path with the Profile Path from chrome://version and --no-headless, using a dedicated profile directory.
Fix Chrome detection if needed
Set --chrome-path or the CHROME_PATH variable if the wrong binary is picked, for example on WSL2.
Run an audit
Ask the agent to run an audit or get Core Web Vitals for your URL on mobile, then on desktop.
Client configuration
{
"mcpServers": {
"lighthouse": {
"command": "npx",
"args": ["@danielsogl/lighthouse-mcp@latest"]
}
}
}Endpoints and ports
npx @danielsogl/lighthouse-mcp@latestStarts the server over stdio.LIGHTHOUSE_LOG_LEVELSet to silent, info or verbose when debugging Chrome launch problems.CHROME_PATHOverrides Chrome auto-detection.What you can do with it
- Core Web VitalsLCP, INP and CLS are reported, with threshold checking.
- Mobile versus desktopOne tool compares both device profiles and supports throttling options.
- Performance budgetsYou pass a budget and the server checks the page against it.
- Structured outputEvery tool declares an outputSchema and returns validated structured content.
- Reference resourcesBuilt-in guides cover Core Web Vitals thresholds, WCAG, SEO and security best practices, budgets and framework tips.
- Analysis promptsPrompts such as analyze-audit-results and compare-audits help the agent interpret results.
- Agentic browsing auditsLighthouse 13 checks how well a page serves AI agents, including WebMCP tools and llms.txt.
- Profile supportA persistent Chrome profile keeps cookies and local storage between runs for logged-in audits.
Tools it exposes
run_audit
Runs a Lighthouse audit. Takes url, with optional categories, device and throttling.
get_accessibility_score
Returns the accessibility score and recommendations for a URL.
get_seo_analysis
Returns SEO analysis and recommendations.
get_performance_score
Returns the overall performance score.
get_core_web_vitals
Returns Core Web Vitals metrics, with an optional threshold.
compare_mobile_desktop
Compares performance across devices.
check_performance_budget
Checks a page against a budget you pass in.
get_lcp_opportunities
Finds LCP optimization opportunities.
find_unused_javascript
Finds unused JavaScript code, with optional minBytes.
analyze_resources
Analyzes the page's resources by type and size.
get_security_audit
Runs a security audit with optional checks.
Frequently asked questions
What is the Lighthouse MCP server?
It is an MCP server that lets an AI agent run Google Lighthouse audits. The agent supplies a URL and receives structured results for performance, accessibility, SEO and security, along with Core Web Vitals, resource analysis and optimization recommendations.
How do I install the Lighthouse MCP server?
Add a server entry with command npx and args @danielsogl/lighthouse-mcp@latest to your MCP client config, then restart. You need Node.js 22.0.0 or newer and Chrome or Chromium. The README also shows one-click installs for VS Code and Cursor.
Which tools does the Lighthouse MCP server have?
The server has eleven tools: run_audit, get_accessibility_score, get_seo_analysis, get_performance_score, get_core_web_vitals, compare_mobile_desktop, check_performance_budget, get_lcp_opportunities, find_unused_javascript, analyze_resources and get_security_audit. It also ships analysis and strategy prompts, plus reference resources such as Core Web Vitals thresholds and WCAG guidelines.
Can it audit pages behind a login?
Yes, with a persistent Chrome profile. Launch with --profile-path set to the Profile Path from chrome://version and --no-headless, log in once, and cookies persist between runs. Chrome needs a dedicated user data directory for remote debugging, not the default one.
Does the Lighthouse MCP server check Core Web Vitals?
Yes. get_core_web_vitals reports LCP, INP and CLS with optional thresholds, and get_lcp_opportunities points at ways to improve LCP. These are Lighthouse lab results, so treat them as a test run rather than field data.
What do I do if Chrome fails to launch?
Set LIGHTHOUSE_LOG_LEVEL to verbose to see Lighthouse output. If the wrong binary is chosen, such as Windows Chrome on WSL2, pass --chrome-path or set CHROME_PATH to the Linux binary explicitly.
Can it audit SEO as well as speed?
Yes. get_seo_analysis returns SEO findings and recommendations, and run_audit accepts a list of categories to include. For a skill that applies fixes, see the web SEO skill and the web quality audit skill.
Lighthouse MCP or Chrome DevTools MCP?
This server focuses on scored Lighthouse audits. Chrome DevTools MCP drives and inspects a live browser. Many teams use one to measure and the other to debug, since the two answer different questions.


