Mozaika MCP server
design.mozaika/mozaikamozaika.designstreamable-httpv1.1.0Scanned
Measured design systems decoded from 588 real products, for coding agents.
23 tools: 2 change things and 21 only read.
- Reads private data · 0
- Reads text others write · 3
- Can send data out · 0
- 2 findings
Tools listed without sign-in. Remote posture
With Anlyon
2 tools here change things. Govern them.
- Anlyon lists the tools and adopts none.
- Each tool you adopt gets approvals and receipts.
- Your agent never holds the key.
A tool your agent calls directly is not routed through Anlyon and is not governed by it. Anlyon Radar only lists and labels, and these labels are the same either way.
What the Mozaika MCP server can do
The Mozaika MCP server is published by mozaika.design, a domain the MCP Registry verified. It has 23 tools. It can create or change data (2 tools: audit_code and generate_asset), but cannot delete data. 21 tools only read.
3 tools bring in text other people can write (get_screen, get_web_benchmark and list_design_changes).
The fixed checks matched 2 findings (long-description and annotation-conflict). Each one comes with the text it matched.
Its remote endpoint lists its tools without sign-in.
Last scanned Nov 14, 2026, version 1.1.0.
Tools
Grouped by what each tool does, most consequential first. Open one for its reasons and full description; descriptions are the publisher's own words.
23 tools
Changes things 2Claude Code: ask
audit_codeWhen the user says a screen looks generic, cheap or off and you need to know WHY — paste the code and get numbers back.1 finding
Why it is in this group
- The input takes a statement or command to run (input.code), which can change anything it can reach.
- The server marks it read-only (readOnlyHint: true), but an annotation never lowers a class.
Findings
- Warn
The server marks "audit_code" read-only (readOnlyHint: true), and its own name, description or input give the class "write".
What this check looks forMatched text
annotations: {"readOnlyHint":true,"openWorldHint":false}; The input takes a statement or command to run (input.code), which can change anything it can reach.
Description, as the server wrote it When the user says a screen looks generic, cheap or off and you need to know WHY — paste the code and get numbers back. This is the one tool here that reads YOUR work instead of someone else's product. Give it the component's CSS, or the JSX/HTML with its class attributes (Tailwind utilities are read on the default scale), or both. It measures what the paste actually contains — type ladder, spacing grid, radii, transition timing, container width — grades each against hundreds of live-decoded real products, and returns findings worst-first, each with the value, its percentile, the corpus median and the change to make. It only ever reports what it could genuinely read, and lists what it could not: it sees the source, not the rendered screen, so colour contrast, hover/focus states and runtime-resolved variables are out of reach — for a public URL, get_score(domain) reads those from the live DOM instead. Free and unmetered. Args: code: the CSS and/or markup to audit. Paste the real thing, not a summary.
Annotations the server declares
- readOnlyHint: true
- openWorldHint: false
generate_assetGenerate ON-BRAND icons with AI: subjects (1-12 short nouns, e.g.No findings
Why it is in this group
- The name "generate_asset" says "generate": a word for a change.
Description, as the server wrote it Generate ON-BRAND icons with AI: subjects (1-12 short nouns, e.g. ["settings gear", "credit card"]) rendered in ONE consistent style. style_from is either a decoded domain ("stripe.com" — the icons match that brand's MEASURED style: accents, stroke, corners) or a house style key ("skeuomorph"). INCLUDED with your account — free accounts get a real daily allowance (enough for a full set), Pro/Lifetime a high one; QA-failed images never count. Returns image_url per subject (1024px transparent PNG) + a zip link.
Only reads 21Show 21 read-only tools
compare_componentsSee how the BEST products each build the same UI atom — one call, cross-product, each with its anatomy read from the live DOM.No findings
Why it is in this group
- The name "compare_components" says "compare": a word for reading.
Description, as the server wrote it See how the BEST products each build the same UI atom — one call, cross-product, each with its anatomy read from the live DOM. Use before building any component: compare_components("Button") returns a ranked panel (one per product) of real buttons, each with measured padding / radius / border / shadow / weight / hover — so you see the real divergence (Linear's pill+shadow vs Vercel's shadowless pill vs Supabase's 6px vs Mercury's 32px) instead of guessing. Args: component_type: e.g. "Button", "Navigation", "Card", "Pricing Card", "Input", "Toggle". industry: optional filter, e.g. "Dev Tools", "Fintech". scheme: optional "dark" or "light" (matches the component's measured background). limit: panel size (1-12, default 8). Returns available component types if the requested one has no matches.
Annotations the server declares
- readOnlyHint: true
- openWorldHint: false
compare_recipesSee how the BEST products each build the same hard pattern — one call, cross-product.No findings
Why it is in this group
- The name "compare_recipes" says "compare": a word for reading.
Description, as the server wrote it See how the BEST products each build the same hard pattern — one call, cross-product. Use before building any complex pattern: compare_recipes("Command Palette") returns a ranked panel (one per product), each entry carrying the at-a-glance layer you pick a reference by — overlay radius, whether it ships a shadow, backdrop filter, the open-motion string and the names of the captured states — so you see that Vercel animates the open where Supabase blurs the backdrop, instead of guessing at the invisible motion/state layer. This is the comparison view; call get_recipe(site, recipe_type) on the one you choose for its full measured anatomy tree, easings, state captures and video. Args: recipe_type: e.g. "Command Palette", "Pricing Table", "Toast", "Data Table", "Multi-step Form". industry: optional filter, e.g. "Dev Tools", "Fintech". scheme: optional "dark" or "light" (matches the recipe's measured overlay background). limit: panel size (1-12, default 8). Returns available recipe types if the requested one has no matches.
Annotations the server declares
- readOnlyHint: true
- openWorldHint: false
compare_sectionsSee how the BEST products each solve the same section — one call, cross-product.No findings
Why it is in this group
- The name "compare_sections" says "compare": a word for reading.
Description, as the server wrote it See how the BEST products each solve the same section — one call, cross-product. Use before designing any section: e.g. compare_sections("Pricing / Plans") returns a ranked panel (one per product) of real pricing sections, each with its reference image_url, pixel-measured spec (bg/text/accents/contrast/scheme/columns/alignment/ whitespace) and the product's core design tokens. Args: section_type: e.g. "Hero", "Pricing / Plans", "Testimonial / Social Proof", "Logo Wall", "Feature", "CTA / Sign-up", "Footer", "FAQ", "Stats / Metrics". industry: optional filter, e.g. "AI Tool", "Fintech", "Dev Tools". scheme: optional "dark" or "light" (matches the section's measured background). limit: panel size (1-12, default 8). Returns available section types if the requested one has no matches.
Annotations the server declares
- readOnlyHint: true
- openWorldHint: false
get_asset_packGet a ready-made AI-generated icon pack (12 consistent icons) — e.g.No findings
Why it is in this group
- The name "get_asset_pack" says "get": a word for reading.
Description, as the server wrote it Get a ready-made AI-generated icon pack (12 consistent icons) — e.g. a house style like 'clay-starter' / 'line-minimal-starter', or a pack generated in the MEASURED style of a decoded brand. Free and unmetered. Each image is a 1024px transparent PNG you can download and use directly (full commercial rights). Unknown slug → lists available packs.
Annotations the server declares
- readOnlyHint: true
- openWorldHint: false
get_componentGet one product's UI component fully specified — its anatomy read from the LIVE DOM (exact padding / height / border_radius / border / box_shadow / font_weight / letter_spacing / t…No findings
Why it is in this group
- The name "get_component" says "get": a word for reading.
Description, as the server wrote it Get one product's UI component fully specified — its anatomy read from the LIVE DOM (exact padding / height / border_radius / border / box_shadow / font_weight / letter_spacing / transition + the real :hover state) plus a reference image_url and the product's design tokens, in a single call. Use this for "build a <component> like <product>", e.g. get_component("Linear", "Button"). This is measured, not guessed — data no model has seen. Returns the available component types if the requested one isn't found.
Annotations the server declares
- readOnlyHint: true
- openWorldHint: false
get_design_driftHas this product's design system CHANGED since it was decoded — and which of its numbers are safe to hard-code?No findings
Why it is in this group
- The name "get_design_drift" says "get": a word for reading.
Description, as the server wrote it Has this product's design system CHANGED since it was decoded — and which of its numbers are safe to hard-code? Mozaika re-measures the most-referenced products from the live DOM every night and keeps a dated ledger, so this answers what a screenshot never can: • verdict "held" — nothing moved for N consecutive nights; the spec is still accurate. • verdict "shifted" — a token changed and the new value stuck (with the date and before/after). • verdict "unstable" — a value alternates between nights: a live A/B test, rotating content, or a page that renders differently each run. `do_not_hardcode` lists it. Call this BEFORE building against a cached spec, and before baking any measured value into a token file. Pairs with get_design_system(site) — that gives the spec, this gives its shelf life. Args: domain: e.g. "stripe.com", "linear.app" (bare domain, no scheme). Free.
Annotations the server declares
- readOnlyHint: true
- openWorldHint: false
get_design_historyMeasured design CHANGE HISTORY for a live-decoded domain — the Decode Ledger.No findings
Why it is in this group
- The name "get_design_history" says "get": a word for reading.
Description, as the server wrote it Measured design CHANGE HISTORY for a live-decoded domain — the Decode Ledger. Token-level diffs between deep decodes over time: "radius 4px→8px", "primary hover #4032C8→#0A2540", "motion dominant 150ms→200ms", each dated. Use it to see how a product's design system is EVOLVING (no screenshot library can backfill this). site = a domain ("stripe.com") or product name. Returns first/last decode dates, decode_count and the dated change entries; empty history = measured, stable so far.
Annotations the server declares
- readOnlyHint: true
- openWorldHint: false
get_design_systemGet a product's full, LLM-verified design system so you can match its exact look.No findings
Why it is in this group
- The name "get_design_system" says "get": a word for reading.
Description, as the server wrote it Get a product's full, LLM-verified design system so you can match its exact look. Use this for "design like <product>" (e.g. site="Linear", "Stripe", "Figma"). Returns (default): color_scheme; colors with named roles (background, text, primary, secondary, accent, link, button_bg, button_text); fonts + font_roles; type_scale; spacing; primary/secondary button; framework + personality. All hex normalized. Deep-decoded products additionally include measured button hover/focus states, a shadow elevation scale (card/overlay/subtle), motion durations + easings, the measured spacing scale, the brand's own CSS custom properties (css_vars), and Icon DNA (icons: style outline/filled/duotone/3d, grid, stroke_weight, corner) — all measured from the live page, not guessed. Match them exactly; pass the domain to generate_asset(style_from=...) to strike icons in this exact style. Your own private BYODS design systems (call list_my_design_systems) resolve first. format: leave empty for the raw token dict. Pass "all" to also get paste-ready DESIGN.md / Tailwind v4 / CSS variables / W3C tokens JSON, or a single format name ("tailwind", "css", "design_md", "tokens", "astryx") to get just that text. "astryx" returns a ready Meta-Astryx defineTheme TypeScript file (measured hover/press states + [light,dark] tuples baked in) — save it and run `npx astryx theme build` for production CSS.
Annotations the server declares
- readOnlyHint: true
- openWorldHint: false
get_productPull an entire product as an agent-ready build kit: the full multi-format design system (tokens + DESIGN.md/Tailwind/CSS/JSON), every curated page, every section grouped by type (w…No findings
Why it is in this group
- The name "get_product" says "get": a word for reading.
Description, as the server wrote it Pull an entire product as an agent-ready build kit: the full multi-format design system (tokens + DESIGN.md/Tailwind/CSS/JSON), every curated page, every section grouped by type (with spec + image_url), plus any user flows. Deep-decoded products' tokens also carry measured button states, shadows, motion and the brand's own custom properties. Call this once for "clone/build like <product>" instead of many small calls. Your private BYODS systems resolve first.
Annotations the server declares
- readOnlyHint: true
- openWorldHint: false
get_recipeThe universal BUILD-KIT fetcher — the measured spec + code to reproduce a piece of UI.No findings
Why it is in this group
- The name "get_recipe" says "get": a word for reading.
Description, as the server wrote it The universal BUILD-KIT fetcher — the measured spec + code to reproduce a piece of UI. `recipe_type` selects which library across three families (all agent-ready through this one call): • COMPONENTS (decoded live from a real product's DOM — Mozaika's wedge): "Command Palette", "Dropdown Menu", "Dialog / Modal", "Login", "Data Table", "Onboarding Tour", "Navbar", "Logo Marquee", "Toast", "Date Picker", "Combobox" — returns the anatomy TREE (each node measured), the MOTION (open/close animation + easing a screenshot can't show), every STATE (empty/results/no-results/keyboard-selected), a webm of it running, and the design tokens. • EFFECTS (open-source WebGL hero backgrounds — Apache/MIT): "Hero Effect" → the shader's full config + fps + install command + license/NOTICE. • MOTION (open-source looping showcase templates — MIT): "Motion Showcase" → the template's full parameter surface + the exact Swiper/anime.js/Motion config + install. Use for "build a <thing> like <product>", e.g. get_recipe("Vercel", "Command Palette") or get_recipe("vanta", "Hero Effect"). A complete, uncopyable, measured build kit. Returns the available types if the requested one isn't found.
Annotations the server declares
- readOnlyHint: true
- openWorldHint: false
get_scoreDesign Score: a 0-100 design audit MEASURED from the site's live DOM (real WCAG contrast pairs, detected type ladder, spacing grid, forced hover states, motion) — scored against th…No findings
Why it is in this group
- The name "get_score" says "get": a word for reading.
Description, as the server wrote it Design Score: a 0-100 design audit MEASURED from the site's live DOM (real WCAG contrast pairs, detected type ladder, spacing grid, forced hover states, motion) — scored against the whole measured-web corpus ("top N% of M systems"). Free to read. Use it to audit the site YOU are building or any competitor: returns dimension scores (typography/color/spacing/motion), UI+UX headline scores, plain-language verdicts, the raw measured evidence chips, and a prioritized fix list — each fix anchored to an evidence index (agent-ready: why + how_to you can apply directly to the codebase). Not scored yet (or refresh=true)? A scan starts (~60-90s) using your account email — call get_score again shortly. Full fix payload requires Pro/Lifetime; everyone gets scores, evidence, verdicts and one complete sample fix.
get_screenGet full metadata + image_url for a single screen by slug.Reads text others writeNo findings
Why it is in this group
- The name "get_screen" says "get": a word for reading.
What it can reach
- untrusted-inputThe description's first sentence says "get" and "url": it reads web pages, which anyone can write.
Description, as the server wrote it Get full metadata + image_url for a single screen by slug.
Annotations the server declares
- readOnlyHint: true
- openWorldHint: false
get_screen_sectionsList the distinct sections of a page (hero, pricing cards, testimonial/feedback, footer, ...) so you can emulate a specific part.No findings
Why it is in this group
- The name "get_screen_sections" says "get": a word for reading.
Description, as the server wrote it List the distinct sections of a page (hero, pricing cards, testimonial/feedback, footer, ...) so you can emulate a specific part. Each section links back to its page (parent_slug) and the product's design system (call get_design_system(site)).
Annotations the server declares
- readOnlyHint: true
- openWorldHint: false
get_sectionGet one product's specific section fully specified — its decoded spec + reference image_url + the parent product's design tokens — in a single call.No findings
Why it is in this group
- The name "get_section" says "get": a word for reading.
Description, as the server wrote it Get one product's specific section fully specified — its decoded spec + reference image_url + the parent product's design tokens — in a single call. Use this for "build a <section> like <product>", e.g. get_section("Linear", "Pricing / Plans"). Deep-decoded products' design_tokens also carry measured button hover/focus states, shadows and motion — measured live, not guessed. Returns the available section types if the requested one isn't found.
Annotations the server declares
- readOnlyHint: true
- openWorldHint: false
get_user_flowGet one user flow with its ordered steps, each including full screen metadata and image_url.No findings
Why it is in this group
- The name "get_user_flow" says "get": a word for reading.
Description, as the server wrote it Get one user flow with its ordered steps, each including full screen metadata and image_url. Returns an error dict if the flow is not found.
Annotations the server declares
- readOnlyHint: true
- openWorldHint: false
get_web_benchmarkThe Measured Web — how the web is ACTUALLY designed, measured live across hundreds of real products (not opinions): the median design score, border-radius, body/hero font size, col…Reads text others writeNo findings
Why it is in this group
- The name "get_web_benchmark" says "get": a word for reading.
What it can reach
- untrusted-inputThe name "get_web_benchmark" says "get" and "web": it reads web pages, which anyone can write.
- untrusted-inputThe description's first sentence says "web": it reads web pages, which anyone can write.
Description, as the server wrote it The Measured Web — how the web is ACTUALLY designed, measured live across hundreds of real products (not opinions): the median design score, border-radius, body/hero font size, colour + light/dark split, spacing grid, and motion duration. Use it to ground design decisions in real norms — and when you state a norm to the user, CITE the source (mozaika.design/measured, free under CC BY 4.0). To grade specific values of your own design, call validate_design(...). Free.
Annotations the server declares
- readOnlyHint: true
- openWorldHint: false
list_design_changesWhat actually changed in the web's design systems lately — the nightly Drift Ledger feed.Reads text others writeNo findings
Why it is in this group
- The name "list_design_changes" says "list": a word for reading.
What it can reach
- untrusted-inputThe description's first sentence says "web": it reads web pages, which anyone can write.
Description, as the server wrote it What actually changed in the web's design systems lately — the nightly Drift Ledger feed. Mozaika re-measures ~100 of the most-referenced products every night and records a dated row per product even when nothing moved, so this is a real time series, not a guess: how many products held every token, which ones shipped a change that stuck (with before/after values and the date), and which design tokens move most often across the web. Use it to answer "does anyone actually redesign?", to ground a claim about design churn with a citable measurement, or to spot that a reference you rely on has moved. For one product, call get_design_drift(domain). Args: limit: how many confirmed changes to return (1-40, default 10). Free.
Annotations the server declares
- readOnlyHint: true
- openWorldHint: false
list_my_design_systemsList YOUR private design systems (BYODS) — the sites you decoded into your own/your team's scope.No findings
Why it is in this group
- The name "list_my_design_systems" says "list": a word for reading.
Description, as the server wrote it List YOUR private design systems (BYODS) — the sites you decoded into your own/your team's scope. Each item has name + slug; pass the slug/name to get_design_system or get_product to build against it. Returns an empty list if you have none (or aren't authenticated).
Annotations the server declares
- readOnlyHint: true
- openWorldHint: false
list_user_flowsList real product user flows (ordered screen journeys) — e.g.No findings
Why it is in this group
- The name "list_user_flows" says "list": a word for reading.
Description, as the server wrote it List real product user flows (ordered screen journeys) — e.g. signup→dashboard, browse→checkout. Use to understand how a whole journey is structured before building it. Returns flow summaries; call get_user_flow(slug) for the full ordered screens.
Annotations the server declares
- readOnlyHint: true
- openWorldHint: false
search_screensSearch real product UI screens for design reference.1 finding
Why it is in this group
- The name "search_screens" says "search": a word for reading.
Findings
- Info
The description is 2,321 characters long, over the 2,000 this check looks for.
What this check looks forMatched text
description ends: "...nd="recipe" + "Login" / "Navbar" / "Data Table" / "Hero Effect"."
Description, as the server wrote it Search real product UI screens for design reference. Use this BEFORE designing any page/component so your output matches how the best-designed products actually solve the problem. Returns structured metadata (description, UX patterns, UI elements, colors, palette) plus an image_url. Section/component/recipe hits also carry `measured` and `retina` booleans — prefer measured:true, retina:true references (pixel-measured, high-res). `query` alone works well — the filters below are optional. A value outside the lists is treated as a HINT (it ranks, it does not exclude), and if the filters together match nothing they relax rather than hand you an empty list. So a near-miss costs you nothing; spelling one exactly is simply more precise. Args: query: free text, e.g. "fintech onboarding", "dark dashboard", "Linear". page_type: one of Billing · Camera / Capture · Changelog · Chat / Assistant · Checkout · Dashboard · Detail · Docs · Editor · Empty State · Feed · Integrations · Landing Page · Log In · Map · Onboarding · Paywall · Player · Pricing · Profile & Account · Search & Results · Settings · Sign Up · Stories. ux_pattern: e.g. "Dark Mode", "Filter & Sorting", "Stats / KPIs", "Data Table", "Command Palette", "Multi-step Form", "Sidebar Navigation", "Bento Grid", "Progressive Onboarding", "Empty State", "Kanban", "Master-Detail", "WYSIWYG". industry: one of AI · Analytics · Communication · Consumer · Creator · Data · Design · Dev Tools · E-commerce · Entertainment · Fintech · Health & Fitness · Productivity · Real Estate · Security · Travel & Local. platform: "Web", "iOS" or "Android" (mobile = official store-listing screens). limit: max results (1-40, default 12). kind: "page" (default, whole screens), "section" (page parts), "recipe" (live-decoded composed patterns: Command Palette, Navbar, Login, Data Table, Hero Effect...) or "component" (measured single components). section_type: narrows by type, e.g. kind="section" + "Pricing / Plans" / "Testimonial / Social Proof" / "Hero", or kind="recipe" + "Login" / "Navbar" / "Data Table" / "Hero Effect".
Annotations the server declares
- readOnlyHint: true
- openWorldHint: false
validate_designGrade YOUR design values against the measured corpus (hundreds of real products, decoded live — not opinions).No findings
Why it is in this group
- The name "validate_design" says "validate": a word for reading.
Description, as the server wrote it Grade YOUR design values against the measured corpus (hundreds of real products, decoded live — not opinions). Pass whichever metrics you have; each returns its percentile, the corpus median/p25/p75 and a verdict. Use this to anti-slop-check your own output BEFORE shipping: "body 13px = p6 (median 16px) — too small" or "radius 24px = p97 — much rounder than real products". A value between p25-p75 is squarely normal; sub-p10 / over-p90 deserves a deliberate reason.
Annotations the server declares
- readOnlyHint: true
- openWorldHint: false
This is not a safety rating, and no finding is not a safety guarantee. The groups and findings come from fixed checks of the tool definitions the server returned on the date shown, not from its code or what it does when called. How the checks work.
Exfil surface
Which tools read private data, which read text someone else can write, and which can send data out. One of each is a path a prompt injection can use.
No private-data, untrusted-input, outbound path in this server alone: no tool is tagged private-data or outbound.
- Reads private data0
- Reads files, mail, messages, databases, secrets, calendars or repositories.
- Reads text others write3
- Brings in text someone else can write: web pages, inbound email, messages, issues and comments.
- Can send data out0
- Can carry data to people or systems: it sends, posts or publishes, or takes a recipient, channel or URL.
Other servers in the same agent add their own tools to these paths. Audit your whole setup to see them together.
Remote posture
What anyone on the internet sees at the remote endpoint: an anonymous initialize and tools/list, and the public OAuth metadata.
Looked at Nov 20, 2026, 18:51 UTC, unchanged since Oct 22, 2026. No credentials were sent and no tool was called. What is looked at.
- Reachable
- yes
- HTTP status of an anonymous initialize
- 200
- tools/list served without credentials
- yes
- Tools that change things, listed without credentials
- 2
- Protocol version
- 2025-06-18
- WWW-Authenticate on 401
- not observed
- resource_metadata in WWW-Authenticate
- not observed
- RFC 9728 protected resource metadata
- not observed
- Authorization servers
- not observed
- Legacy HTTP+SSE transport
- no
- Dynamic client registration
- not observed
- Client ID metadata documents
- not observed
- WarnThe whole server
The endpoint served tools/list without credentials, and the list has 2 tools that change things (2 write). Whether a call also works without credentials was not tested: tools are never called.
What this check looks forMatched text
https://mozaika.design/mcp: tools/list returned 23 tools to an anonymous client, including audit_code (write), generate_asset (write)
Findings
Matches of fixed checks, with the text they matched.
- Warnaudit_code
The server marks "audit_code" read-only (readOnlyHint: true), and its own name, description or input give the class "write".
What this check looks forMatched text
annotations: {"readOnlyHint":true,"openWorldHint":false}; The input takes a statement or command to run (input.code), which can change anything it can reach.
- Infosearch_screens
The description is 2,321 characters long, over the 2,000 this check looks for.
What this check looks forMatched text
description ends: "...nd="recipe" + "Login" / "Navbar" / "Data Table" / "Hero Effect"."
History
Changes to its tool definitions and to its remote endpoint, newest first.
No change seen yet. Its tools are rescanned every day, and a remote endpoint is looked at once a week.
Install
Pinned to version 1.1.0. Values in angle brackets are yours to fill in.
claude mcp add --transport http mozaika https://mozaika.design/mcp --header 'Authorization: <your-authorization>'claude mcp add --transport http mozaika https://mozaika.design/mcp --header 'Authorization: <your-authorization>'Run in a terminal.
Claude Code permissions
Allow the reads, ask before a change or a way out, deny the deletes. Paste into .claude/settings.json.
{
"permissions": {
"allow": [
"mcp__mozaika__search_screens",
"mcp__mozaika__get_screen_sections",
"mcp__mozaika__list_user_flows",
"mcp__mozaika__get_user_flow",
"mcp__mozaika__get_screen",
"mcp__mozaika__get_design_system",
"mcp__mozaika__get_design_history",
"mcp__mozaika__validate_design",
"mcp__mozaika__get_web_benchmark",
"mcp__mozaika__get_design_drift",
"mcp__mozaika__list_design_changes",
"mcp__mozaika__get_section",
"mcp__mozaika__compare_sections",
"mcp__mozaika__get_component",
"mcp__mozaika__compare_components",
"mcp__mozaika__get_recipe",
"mcp__mozaika__compare_recipes",
"mcp__mozaika__get_product",
"mcp__mozaika__list_my_design_systems",
"mcp__mozaika__get_asset_pack",
"mcp__mozaika__get_score"
],
"ask": [
"mcp__mozaika__audit_code",
"mcp__mozaika__generate_asset"
],
"deny": []
}
}Tools that only read are allowed, tools that change things ask first, tools that delete or cancel are denied. A read that can send data out (it takes a URL, say) asks first too. It assumes the server is installed as mozaika. Review it before you use it.
Capability tags behind 3 rules
- mcp__mozaika__get_screen untrusted-input
- mcp__mozaika__get_web_benchmark untrusted-input
- mcp__mozaika__list_design_changes untrusted-input
Pin
Keep the decision: fail CI when the tools change, follow every change, or show a badge.
Pin in CI
Fail a build when the tools you reviewed are not the tools the server serves today.
curl -s "https://radar.anlyon.com/api/v1/servers/design.mozaika/mozaika/verify?hash=a72617f0bf8a7b55fa2a876ddb62758dae576ebf7dbdb8a034e86fc314c824a0" | jq -e '.changed == false'Pinned hash a72617f0bf8a, from Nov 14, 2026, 23:18 UTC.
Watch
Get told when its tools change: a new tool, a new description, a new class.
https://radar.anlyon.com/servers/design.mozaika/mozaika/changes.xmlAn Atom feed. Works in any feed reader or Slack's RSS app.
Badge
[](https://radar.anlyon.com/servers/design.mozaika/mozaika)Source: MCP Registry. Synced Oct 21, 2026, 18:11 UTC.
Mozaika MCP server: questions and answers
Answered from the scan above, not from the publisher's description.
Is the Mozaika MCP server safe?
- No scan can promise that. What its definitions show: 23 tools, 2 of which can change, send or delete something, and 2 findings from fixed checks. Allow the reads, ask before each change, and deny the deletes; the permissions block on this page does that.
Can the Mozaika MCP server delete data?
- No tool is classed as destructive: going by their definitions, none of them deletes, cancels or revokes anything.
Can the Mozaika MCP server do things that cannot be undone?
- No tool is classed as irreversible: none of them sends, pays, publishes or merges, going by their definitions.
What tools does the Mozaika MCP server have?
- 23 tools: 2 write, 21 read. audit_code, generate_asset, search_screens, get_screen_sections, list_user_flows, get_user_flow, get_screen, get_design_system and 15 more.
Does the Mozaika MCP server need sign-in or OAuth?
- Its remote endpoint listed its tools to an anonymous client, without sign-in.
Does the Mozaika MCP server send data to outside services?
- None of its tools sends, posts or publishes, or takes a recipient, channel or URL.
Has the Mozaika MCP server changed its tools?
- Not since Anlyon Radar started tracking it. Last scanned Nov 14, 2026.