Pixelcheck
Gives Claude actual browser primitives so it stops asking you for screenshots. Five core operations: see() snapshots a page with DOM, console, and network state; act() executes click and fill sequences; extract() pulls structured data against Zod schemas; judge() scores pages against rubrics; compare() runs A/B tests. Ships with an 18-persona audit preset that runs real Chromium sessions across device classes, locales, and geographies. Local first, so screenshots and business logic stay on your machine. Install via npm, point your MCP config at pixelcheck-mcp, and your agent can navigate, verify, and score frontend changes without you playing screenshot courier.
- GitHub stars
- 6
- tools
- 12
- auth
- Secret needed
- transport
- STDIO
- env var
- 1
- last updated
- Aug 4, 2026
Tools
12 toolsPublic tool metadata for what this MCP can expose to an agent.
audit_urlFull audit pipeline against one URL: agent loop, scoring, JSON + HTML report.
Full audit pipeline against one URL: agent loop, scoring, JSON + HTML report.
No parameter schema in public metadata yet.
explore_urlQuick autonomous run with a free-form goal; no scenario YAML needed.
Quick autonomous run with a free-form goal; no scenario YAML needed.
No parameter schema in public metadata yet.
seeLooks at a URL once and returns DOM summary, screenshot, console errors and an optional natural-language note.
Looks at a URL once and returns DOM summary, screenshot, console errors and an optional natural-language note.
No parameter schema in public metadata yet.
actDrives an action sequence (click / fill / scroll / screenshot / natural-language act / vision note) and returns per-step status, final DOM and screenshot.
Drives an action sequence (click / fill / scroll / screenshot / natural-language act / vision note) and returns per-step status, final DOM and screenshot.
No parameter schema in public metadata yet.
extractReturns a typed payload from a URL shaped by a supplied JSON Schema, plus DOM / console / screenshot.
Returns a typed payload from a URL shaped by a supplied JSON Schema, plus DOM / console / screenshot.
No parameter schema in public metadata yet.
judgeRubric-driven critique of one URL returning per-criterion scores and severity-graded findings.
Rubric-driven critique of one URL returning per-criterion scores and severity-graded findings.
No parameter schema in public metadata yet.
compareA/B comparison of two URLs against the same rubric.
A/B comparison of two URLs against the same rubric.
No parameter schema in public metadata yet.
list_personasDiscover which personas are installed in a project.
Discover which personas are installed in a project.
No parameter schema in public metadata yet.
list_scenariosDiscover which scenarios are installed in a project.
Discover which scenarios are installed in a project.
No parameter schema in public metadata yet.
list_capabilitiesSelf-describes the server: every shipped tool with kind, cost band, side-effects and dependencies, plus env-var table and cache state.
Self-describes the server: every shipped tool with kind, cost band, side-effects and dependencies, plus env-var table and cache state.
No parameter schema in public metadata yet.
calibrate_criticRuns the critic calibration gate against labeled fixtures and returns pass/fail and agreement metrics.
Runs the critic calibration gate against labeled fixtures and returns pass/fail and agreement metrics.
No parameter schema in public metadata yet.
get_last_reportReads the most recent audit's summary JSON from the local history DB.
Reads the most recent audit's summary JSON from the local history DB.
No parameter schema in public metadata yet.
Sponsors
Agent CaffeineSponsoredClose your lid, come back to finished work
SnugSponsoredThe clipboard manager for the AI eraPromote your product hereReach 125k+ visitors monthly
WaxSponsoredBlock anything with supervision, without erasing your iPhonePromote your product hereReach 125k+ visitors monthly
knobsSponsoredSee your site on any phone, any day, in any languageConfiguration
ANTHROPIC_API_KEY*secretAnthropic API key. Required for LLM-backed primitives (judge, extract-with-schema) and the audit_url / explore_url presets. Meta tools (list_personas, list_scenarios, get_last_report) work without it.
Readme
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 · Vendor-agnostic · MIT-licensed · Yours to own.
Details
- Repository
- github.com/xcodethink/pixelcheck
- Package
pixelcheck- Transport
- STDIO
- Auth
- Required
- Registry
- MCP registry, active