The judgment layer for AI agents — 104 tools that audit what your agent builds, hold it to your taste, and cite the rule behind every finding. Claude, Cursor, or any MCP client.
Every other audit holds your agent's work to the field's rules. The Taste Engine holds it to yours—calibrated once, applied on every build.
audit_taste
A verdict, with the rule behind it
Calibrate a profile once—your accent rules, your banned words, the wrong-and-right examples you've corrected—and bind it to a surface. From then on, audit_taste returns a verdict on anything your agent builds: BLOCK, WARN, or PASS.
Cites the rule, not a vibe. Every finding names a rule_id and quotes the exact evidence. Clauses it can't check deterministically land in not_assessed—it never guesses.
Learns from your corrections.label_finding records accept, revise, or reject as precedent; accepted patterns are suppressed next time, so the judge grows toward your eye.
Yours, and only yours. Profiles live locally under ~/.raven/taste/. Nothing leaves your machine.
audit_taste — raven-mcp
$ audit_taste project:'raven-mcp'verdictBLOCKverdict_line 2 blocking findings — fix before ship.
findings
· voice-no-hype "supercharge your workflow" → cut the hype verb
· voice-no-hype "unlock durable growth" → name the outcome plainlysuppressed 1 accepted precedentnot_assessed 1 clause needs a live capturequoted_evidence_exempt{ elements: 1, chars: 512 }
Recorded replay. Taste stores are local and private by design—this is a captured audit_taste run showing the shape of a verdict, not a live call from your browser.
Cinematic builds
A recipe that names its price
When a surface's taste calls for an AI-generated video hero, Raven returns a build recipe—and the recipe names its own cost. It tells your agent which paid tool the technique depends on, to confirm that spend with you first, and to agree a still-photography or licensed-film fallback before running anything. A surface that opted out of AI is never pushed toward a paid dependency.
build_hint — ai-video-hero
build_hint: ai-video-herotechnique AI-generated video hero over a poster frame
chain one hero still → short clips around one
consistent subject → reserve 4K for a single
final shot, then web-compress; autoplay muted + inline
depends_on Higgsfield MCP (Seedance model) — paid, externalcost confirm the spend with the user before running
fallback still photography or licensed film, agreed first
never a surface that opted out of AI → no paid tool
Knowledge Layers
The knowledge behind the judgment
Design, content, brand, research, service, business — every layer your AI is missing, structured for machines to query and apply.
AccessibilityNielsenGestaltLaws of UXMobile UXResponsive
22
Pattern libraries
Patterns
Field-tested UI patterns for signup, pricing, dashboards, forms, navigation, CTAs—plus content patterns (errors, empty states, notifications) and service patterns (blueprinting, handoff).
Landing PagesPricingFormsHandoff
4
Brand voice systems
Content Systems
Voice & tone from Mailchimp, GOV.UK, Shopify Polaris, and Atlassian. Plus eleven UX-writing principles and content patterns for errors, empty states, notifications, and form validation.
MailchimpGOV.UKPolarisAtlassian
6
Metrics frameworks
Research & Data
Qualitative, quantitative, and usability methods with protocols, sample-size guidance, and bias traps. Metrics frameworks: HEART, AARRR, North Star, conversion funnel, RICE, OKRs.
HEARTAARRRUsabilityA/B Tests
14
GOV.UK standard
Service Design
Stickdorn, Shostack, moments of truth, peak-end, human-handoff patterns. Plus an HTML service-blueprint generator with two-actor (HI-loop) mode for customer↔lawyer, patient↔doctor flows.
BlueprintingHandoffHI-loopGOV.UK
7
2026 visual trends
Brand & Visual
Logo usage, gradient systems, imagery, visual hierarchy, brand-as-system thinking—plus a time-stamped 2026 trends file (bento grids, monospace for tone, neon-on-glass, brutalism rebound).
LogoGradientImagery2026 Trends
5
Strategy domains
Business Strategy
Monetization, retention, onboarding, growth, and metrics—the business context that shapes every design decision.
MonetizationRetentionGrowthMetrics
12
Design systems
Tokens
Production design system tokens in W3C DTCG format. Compose tokens across systems, export as CSS variables, Figma Variables, or JSON. Generate custom systems from a brand color.
StripeLinearDTCGFigma
20
Audit tools
Render & Audit
Render any live URL or screen and grade it—contrast, typography, tap targets, responsive visibility, asset integrity. Native SwiftUI and React Native, plus parity and API-contract checks.
WebSwiftUIContrastParity
Seventy Tools
Seventy tools, organized by job
Seventy focused calls, grouped by what they do—know, create, audit, judge. Claude calls them automatically, with no special syntax.
01
Know
Query the design intelligence — principles, patterns, research, and the reference libraries Raven reasons from.
Principles & patterns
get_principles
Design principles matched to your UI context—heuristics, laws, accessibility, color theory
get_pattern
Field-tested patterns with do's, don'ts, and evidence across UI, content, and service-design types
get_checklist
Pre-publish quality checklist for any UI type—forms, dashboards, landing pages, mobile
get_d4d_framework
Design for Delight—customer empathy, hypothesis, and experiment templates
search_knowledge
Full-text search across all principles, patterns, and business strategy
Strategy, metrics & research
get_business_strategy
Monetization, retention, onboarding, growth, and metrics frameworks
get_metrics_framework
HEART, AARRR, North Star, conversion funnel, RICE, OKRs—with examples
get_research_method
Qualitative, quantitative, and usability methods—with protocols and bias traps
Reference libraries
list_design_systems
Browse the registry of 12 design systems with tokens and metadata
get_design_system
Full tokens—colors, typography, spacing, radii, elevation, motion
list_content_systems
Browse brand voice & tone systems—Mailchimp, GOV.UK, Polaris, Atlassian
get_content_system
A brand's full voice—attributes, tone shifts, vocabulary, grammar, content patterns
get_content_principles
UX-writing principles—clarity, active voice, error anatomy, inclusive language
get_content_pattern
Copy recipes for error messages, empty states, notifications, form validation
get_service_pattern
Blueprinting, human handoff, signup-as-service, omnichannel, moments of truth
get_service_standard
The GOV.UK Service Standard—14 points for service-quality assessment
get_brand_principles
Logo usage, gradient rules, imagery, visual hierarchy, brand-as-system
get_brand_trends
2026 visual trends—bento, monospace, neon-on-glass, brutalism, AI imagery
02
Create
Generate production-ready systems and assets — from a brand color, a prompt, or a composition.
Design systems & tokens
generate_design_system
Generate a complete design system from a brand color—export as HTML, CSS, Figma, or SVG
compose_system
Mix tokens from multiple design systems into a custom composition
get_brand_system
Get a complete design system for building an app with branding like any company
Brand & creative
create_brand_profile
Store local brand memory: colors, fonts, tone, audience, constraints, product notes, and asset references
get_brand_profile
Read a saved local creative brand profile for reuse in generation and campaign jobs
list_brand_profiles
List saved creative brand profiles under the local Raven creative workspace
create_character_profile
Create a character or identity reference profile for consistent image and video generation
plan_creative_campaign
Plan a multi-asset campaign across photos, cards, UGC ads, TV spots, social packs, and storyboards
create_generation_job
Create a provider-ready image, video, audio, 3D, campaign, or analysis job payload
get_generation_job
Read a generation job, its provider payload, status, and runner output if execution was enabled
list_generation_jobs
List local creative generation jobs by status, media type, or brand profile
list_creative_models
Browse provider-agnostic model slots for image, video, 3D, audio, character consistency, and analysis
list_creative_presets
Browse presets for product photos, marketplace cards, UGC ads, TV spots, social packs, storyboards, and infographics
register_creative_asset
Register a local path or remote URL as an asset reference; Raven stores metadata, not file bytes
Service design
generate_service_blueprint
Render a service blueprint as HTML—current vs. ideal, with two-actor HI-loop mode
03
Audit
Check the work — web, native, and cross-platform — against the standards, with evidence.
Web
audit_page
Score any HTML page against Raven's design quality standards — typography, accessibility, responsive, tokens
audit_url
Render a live URL at every viewport & theme, then catch cropped images, blank videos, hover white-wash, and hidden-on-mobile content
audit_layout
Evaluate a rendered page's visual rhythm, alignment, and optical balance
audit_contrast
WCAG 2.1 contrast ratios for every text element on a rendered page—AA/AAA pass-fail with delta-to-pass
audit_responsive_visibility
Flag content visible on desktop but hidden on mobile across breakpoints—catches responsive-hiding bugs that only surface on real devices
audit_tap_targets
WCAG 2.5.5 / 44px tap-target audit—a per-element fix table with the exact CSS to reach the minimum, worst-first
audit_typography
Typographic-scale report—detects the modular ratio, off-scale sizes, line-height drift, and bloated weight ladders
audit_content
Per-item UX-writing verdicts—metrics need a number, CTAs stay action-led, prose drops passive voice and jargon, with rewrites
audit_asset_integrity
Detect content sliced off inside a correctly-sized export—luminance variance flags a Figma export that ended mid-form
audit_device_frame
Catch content cropped inside a device mockup—aspect-ratio cover loss, baked-in pan/zoom drift, and edges sliced at the frame
audit_consistency
Audit multiple routes for cross-page consistency—content-container width and hero heading tier against the inferred canonical
audit_video_playback
Render in headless Chromium and check every <video> actually advances—classifies each clip playing, paused, stalled, empty, or error
suggest_contrast_fix
Given failing WCAG pairs, return the minimal color change that clears the target ratio—the smallest foreground shift, with a background alternative
Audit React Native / Expo source—touchable a11y, 44/48pt targets, font scaling, safe areas, dark mode
Cross-platform & contracts
audit_parity
Compare iOS and Android snapshots against named spatial relationships—catches cross-platform layout drift past device-verified claims
audit_contract
Verify a wire contract is identical across iOS, proxy, and Android—flags missing tokens, schemaVersion drift, and prefix-ordering bugs
audit_api_contract
Run adversarial queries against a live endpoint—flags responses that pass shape but are confidently wrong
Scoring
evaluate_design
Score a design description against relevant principles and patterns
score_creative
Score prompts, scripts, and concepts for hook strength, benefit clarity, channel fit, audience fit, and brand fit
score_page
Score a page across 7 categories—structure, type, color, spacing, accessibility, responsive, tokens—each rated 0–10, derived deterministically
04
Judge
The Taste Engine — calibrate a person's design judgment once, then hold every build to it, with the rule behind each verdict.
Calibrate
create_taste_profile
Build a ruleset and precedent corpus from explicit rules or a DESIGN.md doc—stored locally under ~/.raven/taste
get_taste_profile
Read a profile's rules, scopes, voice, and its surface bindings
list_taste_profiles
Browse every taste profile on this machine
get_taste_interview
A kickoff question set built from the profile's own rules—asks how each surface should read before the first design decision
bind_taste_surface
Bind a profile to a surface—scoped overrides, URL hosts, a voice note, and per-dimension design notes
Judge & learn
audit_taste
Judge HTML, text, or a URL against a profile—verdict BLOCK/WARN/PASS, every finding citing a rule_id and its evidence
label_finding
Append a labeled precedent—accept, revise, or reject; accepted patterns suppress that finding in future audits
record_taste_decision
Log a chosen-vs-rejected decision so recurring choices become interview defaults
list_taste_decisions
Review the decision history behind a profile's evolving taste
generate_taste_portrait
Render a bound surface as a self-contained designed page built from its own taste store
05
Meta
Local usage reflection and release registration.
raven_reflect
Summarize your local Raven usage log—tool counts, recurring warnings, gaps
raven_register
Register an email for Raven updates and a direct feedback line to the creator
Watch it work
No Figma file. No designer.
One prompt builds a real SwiftUI app—then Raven audits every screen against its principle and pattern library, flags the issues, and guides the fixes. Recorded live on the iPhone 17 Pro simulator.
get_principles · evaluate_design
Build from a prompt, design-guided
A single prompt scaffolds a full SwiftUI app—Raven's principles steer layout, color, and type from the first line of code.
get_design_system · get_pattern
Design system applied, not guessed
Activity rings, KPI cards, dark-theme elevation—every choice traced to a principle: 44px targets, 60/30/10 color, Miller's Law.
audit_ios_screen · audit_screen
Real screens, audited live
The Workouts tab built and verified end-to-end—progressive disclosure, smart defaults, and 44pt targets checked against the HIG.
get_pattern · audit_typography
Patterns with evidence
Need a dashboard or streak system? Raven returns the pattern that works—do's, don'ts, and the research behind it—before a pixel is drawn.
Raven runs a full accessibility pass—VoiceOver labels, Dynamic Type, contrast ratios, tap-target sizes—and scores every screen.
evaluate_design
Evaluated, with the fix
The finished app scored against Raven's principles and patterns. Raven flags pure-black backgrounds, color-only indicators, and motion gaps—and returns the fix for each.
Real Outputs
One tool. Every industry.
Built entirely with Raven's design intelligence. Click any to explore.
Production tokens from 12 world-class design systems in W3C DTCG format. Platforms, products, and frameworks — all queryable.
Apple HIG
Platform
iOSmacOSpremium
Material Design 3
Platform
Androiddynamic-color
Stripe
Fintech
professionalclean
Linear
Productivity
darkminimal
Airbnb
Consumer
warmtrustworthy
Spotify
Consumer
darkvibrant
Vercel
Developer
monochromeminimal
GitHub Primer
Developer
enterpriseaccessible
Notion
Productivity
warmcontent-first
shadcn/ui
Component Library
tailwindradix
Tailwind CSS
Framework
utility-firstcomprehensive
Supabase
Developer
darkgreen-accent
The Story
Named for the creature that bridges worlds
Across Norse, Celtic, and Native American traditions, the raven carries knowledge from hidden places and brings it into the light.
Odin's ravens Huginn and Muninn—Thought and Memory—fly across all nine realms at dawn and return to whisper everything they've seen. The Morrígan takes raven form to decide fate. Yéil stole the sun from darkness and gave it to everyone.
Raven MCP: design intelligence that used to be locked away, now flying across your work.
Norse
Huginn & Muninn
Thought & Memory—Odin's ravens that fly across all nine realms and return with knowledge
Celtic
The Morrígan
Sovereignty & Prophecy—the triple goddess who takes raven form to decide fate
Tlingit
Yéil the Transformer
Light from Darkness—Raven stole the sun, moon, and stars and gave them to everyone
Open Source
Free. All of it.
Every tool, every token, every principle — no limits, no tiers, no catch. Raven is open source and always will be.
70Tools
129Principles
12Design Systems
22Patterns
Design intelligence for every AI-generated interface — accessibility audits, token systems, evaluation scoring, and more. Install in 30 seconds.
Need something custom?
Custom Design Systems
Get a bespoke Raven-compatible design system built for your brand — tokens, principles, evaluation rules, all tailored to your product.
Everything you need to know about Raven, the design-intelligence MCP server.
What is RavenMCP?
RavenMCP (Raven) is an open-source Model Context Protocol (MCP) server that gives AI agents like Claude design intelligence — the design knowledge once locked in expert heads. It exposes 104 tools locally (npx, Claude Code/Desktop, or Cursor) spanning design principles, UI patterns, design-system tokens, content voice, brand, research methods, service blueprints, business strategy, multi-platform design audits, and a Taste Engine that holds every build to the owner's own design judgment. The public remote endpoint trades local-only tools (Grab, the Taste Engine's file store) for a smaller stateless subset — see the docs for the exact split.
How do I install RavenMCP?
Install Raven in one command: claude mcp add raven -- npx -y raven-mcp. It runs through npx, so there's nothing to clone or build. Full setup instructions are in the documentation.
Is RavenMCP free?
Yes. RavenMCP is 100% free and open source under the Apache 2.0 license. Every tool, token, and principle is included — no tiers, no usage limits, and no account required.
Which AI agents work with RavenMCP?
Raven works with any client that supports the Model Context Protocol — Claude (Claude Code and the Claude desktop app), Cursor, and any other MCP client. Once installed locally, the agent can call all 104 of Raven's tools directly during a conversation; connecting to the public remote endpoint instead trades local-only tools for a smaller stateless subset (see the docs for details).
What platforms can RavenMCP audit?
Raven can audit web pages and layouts, iOS/SwiftUI screens, Android screens, and React Native interfaces — checking them against design principles and UI patterns, and flagging issues with contrast, tap targets, typography, and accessibility.
What design systems does RavenMCP support?
Raven includes production tokens from 12 world-class design systems in W3C DTCG format, all queryable by AI agents. It also exposes 129 design principles and 22 reusable UI patterns.
What is the Taste Engine?
The Taste Engine makes one person's design judgment portable. You calibrate a profile once — your rules, your banned words, the wrong-and-right examples you've corrected — and bind it to a surface. From then on, audit_taste returns a verdict on anything your agent builds: BLOCK, WARN, or PASS, with the rule and the evidence behind every finding. It learns from your corrections through label_finding, and the profile lives locally under ~/.raven/taste — nothing leaves your machine.
Does Raven cost anything to run, or use paid services?
Raven itself is free and open source, and every tool runs locally. Some optional build recipes — like an AI-generated video hero — can chain a paid external tool such as the Higgsfield MCP. Raven never spends on your behalf: the recipe names its dependency, tells the agent to confirm the cost with you first, and to agree a still-photography or licensed-film fallback before running anything.
Get Started
Give your AI design intelligence
Open source. Zero runtime dependencies. Works with Claude Code, Claude Desktop, Cursor — any MCP client.
