MCP.Market

Pixelcheck MCP server

by xcodethink·io.github.xcodethink/pixelcheck·v2.1.0·6 stars
No reviews yet

Real eyes and hands on the web for AI agents. MCP-first. Local-first. Vendor-agnostic.

C68/100grade C
Write a reviewInstall
What users say
No reviews yet
Be the first
Safety scan
C68/100

full report

Adoption
Growing

6 stars38 downloads/wk

Reviews

Write one

Nobody has reviewed Pixelcheck yet.

If you have run it, two minutes of your experience saves the next person an afternoon.

Pixelcheck tools (23, 4 write)

write = sends, deletes, buys or posts

Read from the package source without running it. The installed server may list more.

  • AUDIT_COMPARES_DIR

    Override the artifacts root for `compare` (one subdir per call).

  • AUDIT_MEMORY_DISABLED

    Set to 1/true/yes to bypass agent memory entirely.

  • AUDIT_MEMORY_PATH

    SQLite file backing the agent memory store. Override to isolate per env.

  • AUDIT_PLAN_CACHE_DISABLED

    Set to 1/true/yes to bypass the planner's cache.

  • AUDIT_PLAN_CACHE_PATH

    SQLite file backing the planner's cache.

  • AUDIT_REPORTS_DIR

    Override the reports root used by audit_url / explore_url. Reports DB lives under `<root>/history.db`.

  • LOG_FILE

    Append all log lines to this file as well as stderr.

  • LOG_LEVEL

    pino log level: trace | debug | info | warn | error | fatal. Default info.

  • LOG_PRETTY

    Force pretty-printed logs (1) vs JSON (0). Default: TTY-aware.

  • actwrite action

    Execute a sequence of actions on a URL: goto / click / fill / press / wait / wait_for / scroll / screenshot, plus AI steps `act` (Stagehand natural language) and `note` (one vision call). Returns per-step status, final DOM summary, console errors, and a final screenshot. Engine auto-selects: Stagehand only if any step.type='act'.

  • audit_urlwrite action

    Run a UX audit against a URL using one persona. Returns the audit summary + report path.

  • calibrate_criticwrite action

    Run the critic calibration gate against labeled screenshot fixtures. Returns pass/fail + metrics.

  • compare

    A/B page comparison primitive. Each side is `{url}` or a pre-captured `{capture}`. Default mode `double_blind` judges each side independently with the same rubric (parallel) then runs ONE synthesis vision call seeing both screenshots — 3 vision calls total, free of anchoring bias (commercial UX-review practice). `fast` mode collapses to 1 side-by-side vision call (cheaper, anchored). Built-in rubr

  • diagnose

    Holistic page-health diagnosis (ADR-034 / Phase 0 entrypoint). Captures a URL with eager visual scoring, reads every diagnostics dimension (performance / network / popups / cookies / storage / visual), and produces a commercial-grade structured report: per-finding severity + dimension + confidence + evidence_refs (cited diagnostics paths) + standards_mapping (Core Web Vitals, WCAG, OWASP, GDPR), a

  • doctor

    Diagnose the pixelcheck environment (Node, API key, browser binary, network).

  • explore_urlwrite action

    Send the autonomous agent to explore a URL with a free-form goal. Faster than audit_url; no scenario file required.

  • extract

    Schema-bound structured extraction from a URL. Caller hands us a JSON Schema describing the desired payload shape; the tool runs Stagehand's extract() under the hood and returns matching `data` plus DOM summary, console errors, and a screenshot. Single LLM call per invocation. Single engine: Stagehand. When `schema` is omitted, returns Stagehand's free-form `{ extraction: string }` driven by `inst

  • get_last_report

    Read the most recent audit's summary JSON from the reports history DB.

  • judge

    Single-page rubric-driven critic. Captures a URL (or accepts a pre-captured snapshot) and runs one vision call against the chosen rubric(s). Built-in rubrics: 'aesthetic' (8 criteria — visual hierarchy, typography, alignment, contrast, spacing, polish, density, brand cohesion) and 'dark_pattern' (12 criteria — forced continuity, hidden costs, pre-selected options, fake urgency, confirmshaming, obs

  • list_capabilities

    Self-describe the MCP server. Returns every shipped tool with its kind, input schema, result schema title, cacheability, static cost-estimate band, side-effects, and dependency declarations; plus the public env-var table and live state of the result cache. Pure introspection — no LLM, no browser, no probe of secret presence. Call this first to plan which tools to use and budget for them.

  • list_personas

    List all personas available in the project's personas/ directory.

  • list_scenarios

    List all scenarios available in the project's scenarios/ directory.

  • see

    Look at a URL once and return DOM summary + screenshot + console errors + an optional vision note. Lightweight primitive: 0 LLM cost when goal is omitted, ~1 vision call when set. Faster than audit_url / explore_url for one-shot inspection.

Public scan report

scanner v0.1.9 · 2026-09-24 · same rubric, same numbers if you re-run it

2 medium
  • Code scan247 source files scanned15/25
  • –Live reliabilityno gateway calls yet and no remote to proben/a
  • –Tool poisoningtools not inspected (local package is not executed); not countedn/a
  • Auth qualitystatic API keys via environment variables6/15
  • Maintenancelast push 3 days ago15/15
  • Maintainer identityregistry namespace matches repository owner; GitHub account older than a year8/10

Findings (2)

  • mediumnpm install lifecycle script presentinstall.script
  • mediumnpm install lifecycle script presentinstall.script
    package.json: …ly && node scripts/chmod-bins.mjs", "postinstall": "node scripts/postinstall.mjs", "pr…
Overall 68/100. Components that don't apply are left out of the denominator. Any critical finding is an F.RubricAppeal a findingJSON

How Pixelcheck compares

Against every graded MCP server, from our own scans and uptime checks. Updated after each scan run.

Safety score
68/100
higher than 30% of 48,255 servers
Tools
23
the typical server has 7
Tools that change things
17%
typical: 0%. Fewer write tools means less an agent can break
GitHub stars
6
more than 85% of 28,688 servers
Downloads a week
38
more than 29% of 12,191 servers with a package

What its tools let an agent do: 19 read · 4 change

Install Pixelcheck in Claude Code, Cursor or VS Code

Runs npx -y pixelcheck on your machine. Read the scan report first; the gateway never runs local packages.

claude mcp add pixelcheck -- npx -y pixelcheck
Add to Cursor

What the publisher says

From the Pixelcheck repository's README, as published. We do not edit it. Read it on GitHub

PixelCheck

Real eyes and hands for the AI agent that's writing your frontend.

Drop-in MCP server. Five browser primitives. Eighteen personas across fifteen countries. Local-first &middot; Vendor-agnostic &middot; MIT-licensed &middot; Yours to own.

Quick Start → &nbsp;&middot;&nbsp; Primitives &nbsp;&middot;&nbsp; MCP Server &nbsp;&middot;&nbsp; Audit Preset &nbsp;&middot;&nbsp; Why Not E2E &nbsp;&middot;&nbsp; Changelog

If PixelCheck helps you, give it a star — it helps others discover the project.

Right now, you're a screenshotting middleman.

Your AI agent is writing 80% of your frontend. It's fast. It's good at code. But it's blind.

  • It writes a button. You open Chrome to check it rendered right. Paste a screenshot back. Ask for the fix.
  • It tweaks the OAuth flow. You log in to verify it didn't silently break. Again. Sixth time this month.
  • It updates the Japanese strings. A user emails: "half the page is in English." You didn't catch it.
  • It rewrites checkout. You walk through it on iPhone, Android, iPad just to feel whether step 3 is confusing.
  • It changes the Arabic layout. RTL didn't propagate. You don't notice for two days.

You become the bridge. The agent has thoughts. You have a browser. The two never meet. Hours of your week, every week, indefinitely.

PixelCheck is the bridge.

A single MCP server. Five primitives. Drop it in once — your agent has eyes and hands.

see(url, opts)              snapshot a page (DOM + screenshot + console + network)
act(url, steps)             execute an action sequence (semantic + selector + Computer Use)
extract(url, schema)        pull structured data matching a Zod / JSON schema
judge(url, rubric)          score a page against a rubric ("is this dark-pattern free?")
compare(a, b, criteria)     A/B comparison of two URLs (incl. blind mode)

Now your agent navigates. Sees rendered HTML. Reads console errors. Clicks. Fills. Judges. Compares. Without ever leaving its workflow — drop into Claude Desktop, Cursor, Cline, Continue, Zed, or Claude Code via four lines in ~/.mcp.json.

Shortened. The full README is on GitHub.

Nothing above is checked by us. What we check is on the safety report.

Pixelcheck: common questions

Is Pixelcheck MCP server safe?
With care: it is graded C, so read the findings first (68/100). Read the Pixelcheck safety report
How do I install Pixelcheck?
It runs on your machine. Copy the Claude Code, Cursor, VS Code or Claude Desktop config from the install section.
Does Pixelcheck need an API key?
Yes. The registry entry asks for ANTHROPIC_API_KEY.
Is Pixelcheck maintained?
The last commit was 44 days ago (2026-09-21). The latest release is v2.1.0.
How does Pixelcheck compare with other MCP servers?
safety score 68/100 (higher than 30% of 48,255 servers); tools 23 (the typical server has 7); tools that change things 17% (typical: 0%. Fewer write tools means less an agent can break). Full comparison.