VibeLens
The VibeLens server provides a single tool that opens a locally running web app in a headless Chromium browser and returns a screenshot, console/network diagnostics, and a sanitized DOM snapshot. It enables AI coding assistants to:
Capture screenshots at desktop (1920×1080), tablet (820×1180 @2x), or mobile (390×844 @2x with touch) viewports.
Capture either the visible viewport or the entire scrollable page.
Collect console errors, warnings, uncaught exceptions, and failed network requests (e.g., 404s).
Obtain a token-optimized DOM snapshot with essential IDs, classes (including Tailwind), role, and ARIA attributes (up to 95% size reduction).
Delay capture by 0–15000ms to allow hydration or data fetching.
Target specific local URLs (e.g., http://localhost:3000/dashboard) with strict validation to localhost and private networks only.
Operate in a read‑only manner, never modifying files or the page.
Click on "Install Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@VibeLenslook at localhost:3000 and find console errors"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
The problem
AI assistants write frontend code well but cannot see what the browser actually rendered. So you become the framebuffer: you open the page, spot the bug, and describe it in prose — "the button is overflowing the card", "spacing looks off on mobile" — and the model guesses which selector you mean.
Three failure modes dominate that loop:
Invisible errors. A hydration mismatch or a 404 asset lives in the console, not in your description.
Selector hallucination. The model edits
.btn-primarywhen the element actually carries.cta-button, because it is recalling its own earlier output instead of reading the DOM.Generated-looking UI. Untouched framework accent, one radius on everything, no type scale, no empty or focus states. It works, and it looks like nobody made it.
Related MCP server: pagelens
The fix
VibeLens is an MCP server that closes the loop. One tool call returns a screenshot, the console and network diagnostics, and a token-optimized DOM snapshot — so the model reasons from evidence instead of memory. The Claude Code plugin adds twelve skills on top, including a design-craft set aimed squarely at the third problem.
Every frame is real output. The screenshots are the tool's own JPEG payload and the console and DOM lines are its own text payload — regenerate them with npm run assets:gif.
What comes back | Why it matters |
Screenshot (JPEG) | The model can see layout, spacing, colour, overflow and hierarchy. |
Console + network errors | Catches hydration errors, uncaught exceptions and 404 assets a screenshot cannot show. |
Sanitized DOM snapshot | Real ids and Tailwind/CSS classes, so fixes target selectors that exist. |
How it works
Install
Claude Code — install the plugin and you get the MCP server plus twelve skills, four subagents and two advisory hooks:
claude plugin marketplace add soumyachk101/VibeLens
claude plugin install vibelens@vibelensEvery other IDE — VibeLens is a plain MCP server on npm. There is no global
install step; npx fetches it on first run:
npx playwright install chromium # one-time, ~95 MBThen paste the config block for your IDE from the next section.
Requirements: Node.js 20 or newer (Playwright 1.62 requires it) and Chromium for Playwright.
git clone https://github.com/soumyachk101/VibeLens.git
cd VibeLens
npm install
npx playwright install chromium
npm run build # emits dist/
npm test # 63 tests, includes real browser captures
npm run smoke # end-to-end check over real stdioThen point your IDE at node /absolute/path/to/VibeLens/dist/index.js.
Configure your IDE
VibeLens speaks MCP over stdio, so every MCP-capable client takes the same shape:
{
"mcpServers": {
"vibelens": {
"command": "npx",
"args": ["-y", "mcp-vibelens@1"]
}
}
}As a plugin (recommended — bundles the skills, agents and hooks):
claude plugin marketplace add soumyachk101/VibeLens
claude plugin install vibelens@vibelensOr as a bare MCP server:
claude mcp add vibelens --scope user -- npx -y mcp-vibelens@1 # every project
claude mcp add vibelens --scope project -- npx -y mcp-vibelens@1 # writes .mcp.json for the teamVerify with /mcp in a session — vibelens should be listed as connected.
codex mcp add vibelens -- npx -y mcp-vibelens@1Or edit ~/.codex/config.toml directly. Codex shares this config between the CLI
and the IDE extension, so it only needs doing once:
[mcp_servers.vibelens]
command = "npx"
args = ["-y", "mcp-vibelens@1"]Check with codex mcp list. Codex only supports local stdio servers, which is
exactly what VibeLens is.
Create .cursor/mcp.json in your project, or ~/.cursor/mcp.json for all
projects, with the JSON block above. Then enable vibelens under
Settings → MCP.
Open the MCP config from the UI — three-dot menu in chat → MCP Servers → Manage MCP Servers → View raw config, or Settings → Customizations → Open MCP Config — and paste the same block. Depending on your build the file lives at:
~/.gemini/antigravity/mcp_config.json~/.gemini/config/mcp_config.json(newer builds and the Antigravity CLI)
Using the in-app menu is the safest way to open the right one.
~/.codeium/windsurf/mcp_config.json, same JSON block.
.vscode/mcp.json:
{
"servers": {
"vibelens": { "command": "npx", "args": ["-y", "mcp-vibelens@1"] }
}
}~/Library/Application Support/Claude/claude_desktop_config.json on macOS, same
JSON block. Restart the app afterwards.
The tool
inspect_localhost_ui
Parameter | Type | Default | Description |
| string, required | — | Local URL, e.g. |
|
|
|
|
| number (0–15000) |
| Milliseconds to wait after load, for hydration, animations or data fetching. |
| boolean |
| Capture the whole scrollable page instead of the viewport. |
Returns two content blocks: an image (base64 JPEG, quality 75) and a text
block containing JSON:
{
"summary": {
"url": "http://localhost:3000/health",
"pageTitle": "Service health",
"viewport": "desktop (1920x1080)",
"fullPage": false,
"waitedMs": 1000,
"captureMs": 1253,
"consoleErrors": 1,
"consoleWarnings": 0,
"uncaughtPageErrors": 0,
"failedRequests": 1,
"domTruncated": false
},
"consoleLogs": [
{ "level": "error", "text": "Hydration failed...", "location": "http://localhost:3000/app.js:42:13" }
],
"uncaughtPageErrors": ["TypeError: Cannot read properties of undefined"],
"failedRequests": [
{ "url": "http://localhost:3000/avatar.png", "method": "GET", "failure": "HTTP 404", "status": 404 }
],
"simplifiedDOM": "<body class=\"...\">...</body>"
}It is read-only and local-only. It observes a page; it cannot click, hover,
type, scroll, or reach a public host. That constraint shapes every skill below —
hover and :focus-visible styles are read from the source, not observed.
Prompts that work well
Check localhost:3000 on mobile and tell me what breaks.
Look at localhost:5173/settings — the cards aren't aligned. Fix it.
Does localhost:3000/checkout throw anything in the console?
This page looks AI-generated. Tell me why, then fix the worst of it.
The accent is just default Tailwind blue — give me a real palette.
Compare localhost:3000 on desktop vs mobile and make the nav responsive.What the plugin adds
The raw tool gives the model eyes. The plugin gives it a method — and a set of skills aimed at the thing screenshots alone do not fix: UI that works but looks like nobody designed it.
Debugging
Skill | What it does |
| Capture → read the DOM and diagnostics → locate the source → fix → re-capture to verify. |
| Groups console and network failures by root cause, maps them to source files, fixes by severity. |
| Captures mobile, tablet and desktop; reports only real breakage; fixes the narrowest breakpoint first. |
| Heuristic pass: missing alt, unnamed controls, unlabelled inputs, heading order, contrast, tap targets. |
Design craft — the anti-slop set
Skill | What it does |
| The flagship. Judges whether the page reads as designed, then fixes what gives it away: no type scale, untouched framework accent, one radius everywhere, uniform spacing, missing empty and loading states, invisible focus, pure black and white, filler copy. |
| How many sizes and weights are actually in use, whether they form a scale, measure, line height, tracking at display sizes, tabular numerals, font-loading shift. |
| Is the accent still the framework default; hardcoded hex instead of tokens; pure black and white; a flat neutral ramp; weak contrast; colour as the only signal of state. |
| Spacing-scale consistency, whether spacing groups related content, container width and measure, optical alignment, grid versus flex misuse, magic z-index, overflow. |
| Durations against distance travelled, easing by intent, which properties are animated, reduced-motion handling, missing exit transitions — then a named token set. |
| The state matrix for every control: hover, active, focus-visible, disabled, loading, selected, error — plus feedback timing, validation timing and touch targets. |
| The last 10%: focus rings, selection colour, scroll behaviour, skeletons that match final layout, empty and error states, cursors, icon optical sizing, favicon and title, 404 and offline. |
Verification
Skill | What it does |
| Capture before, change, capture after with identical parameters, report a concrete diff. Different parameters make the comparison meaningless. |
Subagents
Subagent | Delegate when |
| A UI bug needs diagnosing from what the browser actually rendered. Never proposes a fix for a page it has not seen. |
| You want an independent pre-merge look across three viewports. Reports defects with evidence; does not edit code. |
| You want a verdict on whether the UI looks deliberately designed, ranked by perceived-quality impact, with the proving class quoted. |
| You are building new UI. Establishes tokens before components, builds the full state matrix, never ships a screen it has not captured. |
Advisory hooks
Both are PostToolUse, both are advisory only — they print one line and always
exit 0, so they can never block a tool call or edit a file.
Hook | Fires when |
Unverified change | A |
Raw values | That file contains a six-digit hex, |
Every skill is model-invoked as well as slash-invoked. Saying "this looks
AI-generated" reaches design-review; "the modal just snaps open" reaches
motion-system; "prove that fix worked" reaches before-after.
The design knowledge base
The design skills stay short by citing a shared reference instead of restating it. It is written for an agent mid-task: tables, thresholds and code, no essays.
Document | Contents |
17 recognisable tells of machine-generated UI — the tell, why it reads as generated, and the concrete correction. | |
Modular scales, line height by size, measure in | |
Deriving a palette instead of picking colours, why | |
A 4px scale, spacing as a proximity signal, optical vs mathematical alignment, grid vs flex, container queries, | |
Duration bands by distance, easing by intent, compositor-safe properties, |
Start at docs/design/README.md for the index and the working order.
How it protects your context window
A raw document.body.outerHTML from a modern app is mostly noise. VibeLens
sanitizes it inside the browser, before it ever reaches the model.
Reproduce that measurement yourself with npm run assets:measure.
Removed:
<script>,<style>,<noscript>,<template>,<link>,<meta>,<base>,<title>and HTML comments.Collapsed to an empty placeholder:
<svg>,<canvas>,<iframe>,<video>,<audio>,<object>,<embed>,<map>,<picture>— the box still appears in the tree because it affects layout, but its internals are gone.Attributes kept:
id,class,role,aria-*,data-testid/-test/-cy/-qa, plus form and table structure attributes.Attributes shortened:
data:URIs becomedata:image/png[stripped]; anything over 300 chars is truncated; inlinestyleover 120 chars is cut.Text nodes: whitespace collapsed, each node capped at 160 chars.
Hard cap: 20,000 characters, with an explicit
<!-- VibeLens: DOM truncated ... -->marker so the model knows the tree is partial.
Architecture
Module-by-module detail, the request lifecycle and the resource model live in docs/ARCHITECTURE.md. The reasoning behind each significant choice is recorded as an ADR in docs/adr/:
ADR | Decision |
MCP over stdio | |
Playwright over Puppeteer | |
An allowlist, not a denylist | |
One tool, not many | |
The plugin ships from a subdirectory | |
JPEG screenshots and DOM truncation |
Security
VibeLens drives a real browser on your machine, so an unvalidated URL would be a server-side request forgery (SSRF) primitive. Every URL is checked before Chromium launches.
Allowed: localhost, *.localhost, 127.0.0.0/8, 0.0.0.0, 10.0.0.0/8,
172.16.0.0/12, 192.168.0.0/16, [::1], IPv6 unique-local (fc00::/7) and
link-local (fe80::/10).
Blocked:
Every public hostname and IP address.
Any DNS name that is not
localhost-based — resolving one invites DNS rebinding, so hostnames are refused rather than looked up.Cloud instance-metadata endpoints (
169.254.169.254,169.254.170.2,fd00:ec2::254,100.100.100.200) and all of IPv4 link-local169.254.0.0/16.Non-HTTP schemes (
file:,ftp:,javascript:, ...).URLs containing credentials (
http://user:pass@...).
Two further points worth knowing:
The tool is annotated
readOnlyHint— it observes a page, it never modifies your project, and it cannot click, type or scroll.Page content is untrusted input. Text scraped from a rendered page could contain instructions aimed at your assistant. Treat the DOM snapshot as data, never as direction.
To report a vulnerability, see SECURITY.md.
Each call launches an ephemeral Chromium and closes it in a finally block, so a
failed capture cannot leave a zombie browser behind. This matters on Apple
Silicon laptops where every stray Chromium costs hundreds of MB of RSS.
Troubleshooting
Message | Cause and fix |
| Run |
| The dev server isn't running, or is on another port. |
| The URL isn't local. VibeLens only inspects localhost and private-network addresses. |
| Chromium refuses a fixed port list (1, 7, 69, 79, 6000, 6666...). Use 3000, 5173, 8080. |
| Page never finished loading. Retry, or raise |
Screenshot is blank | The app hadn't hydrated — raise |
Tool not listed in the IDE | Check the config path, then restart the IDE. Logs go to stderr prefixed |
The long-form version, including per-IDE diagnosis, is in docs/TROUBLESHOOTING.md. Common questions are answered in docs/FAQ.md.
Development
npm run dev # run the server from source over stdio
npm run typecheck # tsc --noEmit
npm test # vitest: security, DOM, capture e2e, MCP protocol
npm run build # emit dist/
npm run smoke # spawn dist/ and exercise it over real stdio
npm run validate:manifests # npm + plugin + marketplace + docs consistency
npm run validate:plugin # claude plugin validate for both manifests
npm run assets:measure # reproduce the DOM-reduction figure
npm run assets:gif # regenerate the README animation from real captures
npm run site:build # build the documentation site into site-dist/
npm run site:check # link-check it, then inspect it with VibeLens itselfRepository map
src/
├── index.ts executable entrypoint: wires the server to stdio
├── server.ts MCP server + inspect_localhost_ui registration
├── browser.ts Playwright capture engine
├── dom.ts in-page DOM sanitizer + truncation
├── security.ts SSRF guard (URL allowlist)
└── types.ts shared types, viewport presets, payload limits
tests/ vitest suites, incl. real Chromium captures
├── security.test.ts 44 allow/block cases
├── dom.test.ts truncation boundaries
├── capture.test.ts end-to-end against a noisy fixture page
├── server.test.ts real MCP client over InMemoryTransport
└── fixture-server.ts the deliberately broken fixture page
plugin/ what Claude Code installs — no package.json, on purpose
├── .claude-plugin/plugin.json
├── .mcp.json launches npx -y mcp-vibelens@1
├── skills/ 12 skills: 4 debugging · 7 design craft · 1 verification
├── agents/ ui-debugger · ui-reviewer · design-reviewer · frontend-builder
└── hooks/ two advisory PostToolUse reminders
docs/
├── ARCHITECTURE.md modules, lifecycle, resource and token model
├── design/ the anti-slop, type, colour, layout and motion rules
├── adr/ six architecture decision records
├── PRD-TRD.md product + technical requirements
├── TROUBLESHOOTING.md every error code, per-IDE diagnosis
└── FAQ.md honest answers, including the limitations
site/ hand-authored design system for the docs site
├── styles.css tokens, type scale, themes; no framework
└── app.js theme toggle, copy buttons, TOC highlight
scripts/
├── smoke.mjs spawns dist/ and drives it over real stdio
├── validate-manifests.mjs keeps npm + plugin + marketplace + docs in sync
├── site/ static site generator + its own link checker
└── assets/ regenerates the README animation and measurements
.claude-plugin/marketplace.json one-plugin marketplace catalog
.github/ CI, tag-driven release, issue forms, CODEOWNERSContributing
Start with CONTRIBUTING.md — it lists the seven invariants that must not break (chief among them: never write to stdout, it is the MCP transport) and the verification checklist every PR must pass. Project guidance for AI agents working in this repo lives in CLAUDE.md.
Release process: RELEASE.md. Version history: CHANGELOG.md. Conduct: CODE_OF_CONDUCT.md.
Roadmap
Element-scoped capture (
selector: "#navbar")Interaction before capture (click, hover, fill, scroll) — would let the state-matrix skills observe hover and focus instead of reading them
Multi-viewport diffing in a single call
Full network waterfall for failed API calls
Accessibility-tree output alongside the DOM
Ideas and votes belong in Discussions or a feature request.
Free forever, MIT licensed — see LICENSE. No paid tier, no hosted service, no telemetry.
If VibeLens saved you a round of "no, the other button", a star helps other people find it.
Maintenance
Resources
Unclaimed servers have limited discoverability.
Looking for Admin?
If you are the server author, to access and configure the admin panel.
Tools
Related MCP Servers
- AlicenseAqualityDmaintenanceEnables AI coding assistants to control and inspect a live Chrome browser for automation, debugging, performance analysis, and screenshot capture through Chrome DevTools.Last updated261,395,4223Apache 2.0
- Alicense-qualityDmaintenanceEnables AI coding agents to visually interact with frontend apps by taking screenshots, clicking elements, reading console logs, and performing visual diffs.Last updated3MIT
- Alicense-qualityDmaintenanceEnables AI assistants to take screenshots of localhost development servers for visual verification of UI changes, reducing manual debugging cycles.Last updated50MIT

Inspector Jakeofficial
Alicense-qualityFmaintenanceConnects AI assistants to Chrome DevTools, enabling page inspection via ARIA trees, screenshots, console logs, network monitoring, and element interaction through point-and-click.Last updated27MIT
Related MCP Connectors
Live browser debugging for AI assistants — DOM, console, network via MCP.
Give AI coding agents access to your Vynix visual feedback, bug reports, and AI diagnosis.
Generate images, GIFs, and PDFs from HTML, URLs, or templates — from your AI agent.
Latest Blog Posts
- Who's Calling? MCP Hosts Are an Identity Blind Spot (And the Spec Knows It)By Om-Shree-0709 on .mcpAgent IdentityOAuth 2.1
- Your AI Chatbot Just Exposed Your CEO's Salary to an InternBy Om-Shree-0709 on .Agent IdentityMCP SecurityOAuth Delegation
- Why MCP Servers Need Execution Sandboxing (And Why Your Current Stack Isn't Enough)By Om-Shree-0709 on .Agentic AiPrompt InjectionWebAssembly
MCP directory API
We provide all the information about MCP servers via our MCP API.
curl -X GET 'https://glama.ai/api/mcp/v1/servers/soumyachk101/VibeLens'
If you have feedback or need assistance with the MCP directory API, please join our Discord server