Skip to main content
Glama
aura-components-mcp

aura-components-mcp

I got tired of opening twenty Aura tabs every time I started a landing page. So I built the MCP server I wanted: every free thing on aura.build, one stdio call away, no account, no key, no browser.


Contents

v1.7.3 — hardening release, no new tools, no new dependencies. The whole award layer now fetches through one guarded path (scheme/host/DNS policy, re-checked redirects, size-capped bodies) plus the IPv6 parser fix that had mis-blocked public AAAA/dual-stack hosts. Plan-based index lifecycle (rebuild continues, resume finishes, refresh forces), drift-guarded winner baselines that an empty listing can't wipe, normalized URL watches, and artifact-honest capture/motion. 49 new hermetic tests lock it in: 275 hermetic + 45 live checks, all green.

v1.7.1 — bug-fix release, no new tools. Central free-only enforcement: Pro components/assets by direct ID are now refused (-32602, fail-closed on ambiguous entitlement) instead of returned in full — search, install, bundle, scaffold, and related all inherit it. Search cache isolates explicit theme (dark/light no longer share entries). Bundles honor all 8 refs end-to-end with dropped_refs disclosure. New min_views_filtered_out count (total stays upstream). Full icon outages surface as upstream errors, not misses. Upstream hardening: malformed rows disclosed via malformed_dropped (single fetch fails loud), non-429 4xx never retried. Node floor raised to maintained releases (engines >=22, CI matrix 22.x/24.x). Dead code removed, scaffold merge extracted, verify extended.

v1.7.0 — 29 tools: 5 pure aura_inspo_* mappers (recommend/components/reference/system/compare) bridge Inspo evidence to Aura calls with zero extra deps — strict macro gate (need_macro), palette priority brand > Inspo > Aura, enforced tokens-first, mandated image srcset plan, fast preview-URL look gate — plus 2 pure aura_docs_* helpers (plan/check) wiring Context7 fresh docs for tailwind/gsap/lenis/iconify, with docs_stamp on every install/scaffold and a look-preserving conflict rule (fresh API wins, look stays identical) — plus live aura_icon (Font Awesome Free + HackerNoon pixel SVGs, auto-fallback, CC BY 4.0 credit always) — plus aura_version (hourly npm check, prefer-latest notice, verdict inside every aura_status). aura_prompt now asks 6 questions (+macrostructure, +vibe) and leads Pass 2 with inspo.recommend. aura_scaffold_page refuses explicit Pro ids (pro_filtered_out). Build hardened for CRLF checkouts.

v1.6.3 — 20 tools. New aura_prompt: need-based prompt in 2 passes (draft + 4 questions, ask user, then copy-paste final). Honest search: zero-overlap fallbacks return total:0 + no_match + refused_fallback + suggestions instead of masked popular rows; exact_match:true skips fallbacks; recommend auto-retries once then fills from editorial picks; theme reorder-not-drop; design-systems index content/preview_html; kind-aware skill recall; slim lists by default (slim:false opts out); full:true one-shot bodies; scaffold takes ["732","1364"] / [{id}]; status.freshness per surface; per-surface sort:'trending' (no dead trending tool). Search facets equal detail facets, skills deduped, token starter + preview-mode warnings, legal asset plans, scoped CSS + chunk concat plans, bulk-fetch, one-turn scaffold builds.

Point any MCP client at dist/server.js and your agent can search 2,495 free components, read 192 agent skills in full, pull 30,688 assets, and apply 725 DESIGN.md systems. It answers in seconds because there is nothing to log into and almost nothing to download: one bundled JS file, zero dependencies.

Built by swastiksingh-dev. Catalogue content belongs to Aura (aura.build, by Meng To / DesignCode). This project is not affiliated with Aura.


Related MCP server: you-design-mcp

Why I built this instead of using the official MCP

Aura already ships an official MCP at https://mcp.aura.build/mcp. It is good at what it does: your Canvases, your projects, publishing, all behind OAuth. This server does the other half. It covers the public catalogue, and it skips the parts that slow an agent down: no OAuth dance, no account, no per-project scopes. A few things the official one does not do:

  • aura_install_component tells the agent exactly how to paste a component: which CDN scripts it needs, which fonts it references, which file to put it in.

  • aura_use_design_system hands back a :root token starter plus the copy order (tokens first, markup second).

  • aura_categories shows where the free catalogue is deepest, with live counts; per-surface sort:'trending' shows what is new.

  • aura_recommend builds a starter kit across all four surfaces in one parallel call.

  • Everything defaults to free. Pro rows never sneak into results.


What is inside

  • 2,495 free components with HTML/Tailwind source, preview images, tags, and author credit

  • 192 agent skills with full SKILL.md bodies (GSAP, Tailwind v4, Anime.js, copywriting, and more)

  • 30,688 images and clips with direct CDN URLs at multiple widths

  • 725 DESIGN.md systems with tokens, type rules, layout notes, and preview HTML

  • 35 tools over stdio (29 catalogue + 6 award-intelligence), one bundle, 60-second LRU cache (300 keys) with request coalescing, author memo, category-count cache, retries with jittered backoff

  • Award corpus: file-based JSON index (~/.aura-mcp/awards) with atomic writes, resume checkpoints, winner baselines, watches; polite ~1 req/s Awwwards fetching with distinct 404/403+429/5xx/timeout handling

  • Honest search: AND-phrase misses retry as OR-tokens (2 passes) plus a labeled safety net — but zero title-overlap fallbacks are refused as total:0 + no_match + suggested_queries instead of masking with popular rows; exact_match:true skips fallbacks entirely

  • Agent accelerators: aura_prompt (2-pass prompt with user confirmation), aura_bulk_fetch/aura_bundle (8 details/turn, any of 3 kinds), aura_scaffold_page (tokens + system + markup in dependency order), aura_related, aura_install_asset (license-aware)

  • Faceted lists: every component carries facets { theme, weight, code_chars, needsTailwind, needsIcons, fonts }; filter theme: dark|light server-side

Start with aura_status, then aura_search_all. That order matters: status confirms the catalogue is reachable, search_all shows which surface has the best match before you spend calls on details.


Install

Needs Node 22 or newer. No other dependency. Pick one:

# Option A — from source (GitHub)
git clone https://github.com/swastiksingh-dev/Aura-Components-MCP.git
cd Aura-Components-MCP
npm install
npm run build
node ./dist/server.js
# Option B — npm
npm i -g aura-components-mcp
aura-components-mcp
# Option C — npx, no install
npx -y aura-components-mcp@latest

Then register it in your client. Full per-client steps live in SETUP.md — Claude Code, Cursor, Codex, Windsurf, OpenCode, Cline, Roo Code, DeepSeek Harness, generic stdio.

# Claude Code
claude mcp add aura-components -- node ./dist/server.js
// Cursor, Windsurf, and most others (~/.cursor/mcp.json or equivalent)
{ "mcpServers": { "aura-components": { "command": "node", "args": ["./dist/server.js"] } } }
// OpenCode (opencode.json) — see https://opencode.ai/docs/mcp-servers
{ "$schema": "https://opencode.ai/config.json", "mcp": { "aura-components": { "type": "local", "command": ["node", "./dist/server.js"], "enabled": true } } }

Demo

aura_prompt turns one sentence into a confirmed prompt, aura_recommend turns it into a starter kit with links. Full clip in assets/.


The 35 tools

Start with aura_status, then aura_search_all. That order matters: status confirms the catalogue is reachable, search_all shows which surface has the best match before you spend calls on details.

v1.7.2 — 35 tools: the six award-intelligence tools (aura_award_search, aura_award_inspect, aura_design_blueprint, aura_visual_audit, aura_motion_study, aura_award_index) join the 29 catalogue tools — ranked Awwwards references with try_for cues, design DNA with suggested ideas, blueprints with paste-ready tokens + executable next_calls, graceful browserless fallbacks, and a persistent local corpus. See CHANGELOG.md.

Find

Tool

What it returns

aura_status

Reachability plus live free counts. Call it first.

aura_search_components

Free components by text, category, theme: dark|light, sort. Slim rows by default (excerpt + full_fetch, no 12KB code dump; slim:false for full). exact_match:true skips fallbacks (honest empty).

aura_search_skills

Skill metadata (title, description, source repo, views). exact_match:true skips fallbacks.

aura_search_assets

Images and clips by keyword, media type filter included. Slim rows (no 4K URLs until install). exact_match:true skips fallbacks.

aura_search_design_systems

DESIGN.md metadata: title, description, views, author. exact_match:true skips fallbacks.

aura_search_all

All four surfaces in parallel. Never empties (fallback + suggestions). exact_match:true for honest empty.

aura_recommend

A starter kit for a goal sentence, with links and a short rationale.

aura_categories

The 13 component categories with live free counts. Counts overlap (see overlap_note).

aura_related

3 more-like-this per item. Discovery never dead-ends.

Fetch (full bodies)

Tool

What it returns

aura_get_component

One component in full: markup, style block, preview, author, page URL. full:true ignores chunk for the full body.

aura_get_skill

The whole SKILL.md body. This is the one agents actually build from.

aura_get_design_system

Full DESIGN.md content plus preview HTML.

aura_bundle / aura_bulk_fetch

2–8 details in one call (components/design_systems/assets). Per-item errors never fail the batch.

Install & build (tokens first, then markup)

Tool

What it returns

aura_install_component

The same component plus a paste plan: CDN scripts, fonts, steps, file map, chunk concat plan when truncated, css_scope_how.

aura_install_skill

Where to save SKILL.md so the agent can load it, plus upstream link.

aura_install_asset

Legal drop-in: download + preview URLs, license (all-rights-reserved per Aura Terms §4), file path.

aura_use_design_system

Token starter CSS plus the order to apply things in.

aura_scaffold_page

One-turn page build: tokens.css + system preview + markup in order, combined deps + files. components: ["732","1364"] or [{id:"732"}]. Goals with #hex/punctuation are escaped safely; an unmatched system is disclosed via system_note instead of silent empty CSS.

aura_prompt

Need-based prompt in 2 passes: draft + 6 questions (theme, sections, system, assets, macrostructure, vibe), then copy-paste final. Confirm with user first. Pass 2 leads with inspo.recommend + aura_inspo_recommend.

Bridge (Inspo · fresh docs · icons)

Tool

What it returns

aura_inspo_recommend

Brief → Inspo + Aura call plan (pure, no fetch). Strict macro gate: no macrostructure returns need_macro:true. Palette priority brand > Inspo > Aura.

aura_inspo_components

Inspo component filter → Aura category search + scaffold hint (tokens first, weight s|m).

aura_inspo_reference

One Inspo JSX + one Aura component → scoped adapt plan (token swaps, css scope, files).

aura_inspo_system

Inspo tokens vs Aura DESIGN system → A/B/blend + one system_choice question + contrast warning.

aura_inspo_compare

2–4 slim Inspo sites → house style + closest Aura system/component queries.

aura_docs_plan

Fresh-docs lookup plan (pure): library + need → Context7 resolve/query plan (trusted: tailwind, gsap, lenis, iconify) + Aura apply steps.

aura_docs_check

Pasted code vs fresh docs (pure): verdict current|needs-update, look-preserving fixes, pin checks.

aura_icon

One free icon as ready SVG (live): FA Free + HackerNoon pixel lib, auto-fallback, license + credit always, font alternative, 24h cache.

aura_version

Running vs npm-published version (live, hourly cache). update_available + prefer-latest notice. aura_status carries the verdict on every call.

Award intelligence (new in v1.7.2 — 6 tools, zero npm deps)

Search → inspect → blueprint → audit → motion. Core search/index work with no browser; live capture needs Chrome (see setup below).

Tool

Modes

What it returns

aura_award_search

sites / categories / elements / motion

Ranked Awwwards refs (title, source URL, tags, awards, studio, preview link, relevance + reasons). Multiword queries get alias expansion (glassy→glassmorphism, 3d→webgl) plus a theme:dark/light boost — reasons always name your word, never the alias. categories returns real taxonomy from the index/source, filters paged by limit/offset (filtersTotal tells the full size). motion searches stored studies (emptyCorpus:true when none recorded yet). Every response carries source + corpusNote: first run on an empty index seeds from one polite live fetch (source:live); after that everything ranks from the local index until you rebuild/refresh. Empty results carry suggestions; upstream failures carry liveError instead of fake hits.

aura_award_inspect

site_details / site_elements / element / compare_sites

Design DNA (palette, tech, elements, awards, score). Empty Technologies sections are backfilled from listing tags and labeled technologiesInferred:true (never presented as parsed). Rule-based ideas (0–6 by evidence, always status:suggested + observed basis, contract in ideas_contract). Browserless motion evidence in motion_static (scripts, keyframes, scroll cues — labeled static-markup, never runtime). Element media URLs with attribution. 2–3-site comparisons with shared signals, fuzzy element-keyword overlap, tech-family canonicalization, luminance-band agreement, directionHints, and gaps (missing = unknown, never negative proof).

aura_design_blueprint

goal + 2–5 refs

Markdown brief + versioned JSON (spec.version: 1): shared patterns, outliers, palette evidence, one catalogue query PER proposed section (tokens/hero/proof/…/footer, capped at 12 matches — slim rows, not full code dumps), real free Aura matches (IDs + page URLs, free-only enforced, fallback-sourced rows marked (via fallback …) with matchEvidence), requested hexes from your goal synthesized as proposed tokens against the refs' observed bands, executable next_calls (get/scaffold/motion with real gathered IDs), evidence-scored confidence + confidenceWhy, sections, motion/a11y/performance guidance, checklist, and an always-present Evidence gaps section. Catalogue failures surface as unavailable evidence.

aura_visual_audit

capture / structure / compare

Real browser screenshots + live-DOM band maps when Chrome works (source: live-browser, attached via /json/list page targets). Any live-dom failure degrades to labeled static-html estimates (fallbackFrom + fallbackWhy disclosed) instead of erroring. Local builds: pass a file:// path or local file with allowLocal:true for direct file estimates (source: local-file); screenshots still need a served http://127.0.0.1 URL. compare reports section-count/order, height, palette mismatches + manual-judgment items, with an estimateNote when offsets aren't measured. Capture never claims an artifactPath it didn't write (see artifactNote) and prunes old temp captures (newest 20 kept) with a retentionNote when no saveDir is set; a navigation to a blocked/non-http(s) URL after settle aborts with UNSUPPORTED_OPERATION + blockedUrl.

aura_motion_study

inspect / record / search

Live runtime scans (libs, easing vocab, scroll model, ScrollTrigger counts — only what was observed), bounded frame studies (2–16 frames, ≤30s), .webm only when system ffmpeg exists. Stored studies are searchable by library/easing/scroll. record validates frames before encoding and reports framesUsed; temp aura-motion-* dirs older than 24h are pruned with a retentionNote when no saveDir is set.

aura_award_index

status / rebuild / resume / refresh / new_winners / watch_add / watch_list / watch_remove

Persistent corpus in ~/.aura-mcp/awards (AURA_AWARDS_DIR overrides): counts/freshness/lock, polite ~1 req/s crawl bounded by maxPages per call (default 5, max 20). rebuild continues an unfinished cycle (starts a new one after completion), resume finishes pending work, refresh forces a new crawl. Winner baseline-then-delta is drift-guarded (an empty upstream listing reports degraded:true instead of overwriting the baseline). Watches accept studio/tag or canonical /sites/<slug> URLs/slugs (legacy raw URLs still match). Never auto-crawls on startup.

Browser setup (optional, capture/motion only): install Chrome/Chromium, or set AURA_CHROME_PATH to its executable. For .webm motion video, also install ffmpeg or set AURA_FFMPEG_PATH. Without them the six tools still search, inspect (via polite HTML fetch), blueprint, index, and report BROWSER_UNAVAILABLE with setup guidance where a live browser is required. CDP setup failures (refused socket, hung attach) normalize to the same BROWSER_UNAVAILABLE code with a per-operation fallback hint — never a raw CDP timeout.

%%{init: {'theme': 'base', 'themeVariables': {'fontFamily': 'Inter, system-ui, -apple-system, sans-serif', 'fontSize': '15px', 'lineColor': '#ff2a2a', 'edgeLabelBackground': '#0a0a0b'}}}%%
flowchart LR
    S["🔍 aura_award_search<br/>ranked refs + try_for"] --> I["🧬 aura_award_inspect<br/>design DNA + ideas"]
    I --> B["📐 aura_design_blueprint<br/>brief + tokens + next_calls"]
    B --> A["📸 aura_visual_audit<br/>capture / bands / compare"]
    A --> M["🎞️ aura_motion_study<br/>scans + studies"]
    B -.-> X["🗂️ aura_award_index<br/>corpus · winners · watches"]
    X -.-> S
    classDef core fill:#111318,stroke:#8b949e,color:#f5f5f5,stroke-width:2px;
    classDef hot fill:#1c0606,stroke:#ff2a2a,color:#ffffff,stroke-width:2px;
    classDef idx fill:#161a08,stroke:#d4a017,color:#ffffff,stroke-width:2px;
    class S,I,A,M core;
    class B hot;
    class X idx;

The 6-call award loop (fast: details + section queries run concurrently): search → inspect → blueprint → follow its next_calls → audit → motion

Two award servers? If you also run awwwards-mcp, keep both or pick one: aura_award_* is canonical here — it's the only path that maps references to Aura's free catalogue inside aura_design_blueprint. The standalone server can't do that mapping.

Every row carries page_url (open it to see the design) and author where Aura credits one.


Fallback rules (read this once)

  • Per surface, not global. aura_search_all fans out to four independent pipelines (phrase → OR-tokens → safety net → kind-aware refusal). Mixed outcomes are normal and correct: skills may return fallback:or-tokens:... while components return total:0, no_match:true. Read isFallback / no_match on each surface — never assume one flag covers all four.

  • Recommend fills the kit. If components come back no_match, aura_recommend auto-retries once with suggested_queries[0] (retried:true, retried_from:<query>), then falls back to human-grade editorial picks (editorial_fallback:true). The starter kit is never left empty on a text goal.

  • Big code in one call. Search rows are slim excerpts by design. Skip the 3-chunk manual concat: aura_get_component {id, full:true} and aura_scaffold_page {goal, full:true} return complete bodies (33,897-char component verified in one call). Default stays chunked to protect context.

  • Counts are live upstream. aura_status counts mirror the Aura catalogue right now; freshness shows max updated_at per surface. Counts only grow when Aura publishes — a freshness bump with flat counts means metadata moved, not new components.

  • total means upstream matches (for limit/offset paging), not post-filter rows. Local drops are disclosed, never silent: min_views_filtered_out, negative_theme_filtered_out, deduped, bundle dropped_refs.

  • Award errors are labeled, never masked. NOT_FOUND (−32004) = unknown slug; UPSTREAM_ERROR (−32000) = blocked/failing source (blocks are never retried); BROWSER_UNAVAILABLE = install Chrome; INDEX_BUSY = crawl already running (resume continues); INDEX_CORRUPT = file quarantined + rebuilt empty. Empty award results carry suggestions or emptyCorpus:true; failures carry liveError. Award fetches pass a URL policy (http(s) only; private/loopback/link-local/metadata ranges blocked; bounded best-effort DNS check; redirect targets re-checked; bodies size-capped) — allowLocal relaxes loopback only. Refusal means zero rows matched (best-row gate, theme-first ordering kept). Site ideas run 0–6 by evidence (1 on thin evidence is valid) and are always status:suggested with an observed basis. Blueprint confidence is evidence-scored (refs + matched sections + recorded motion + measured structure + fully-parsed tech → low…high, with confidenceWhy). Known limits: taxonomy/element depth follows the polite crawl (maxPages is a per-call budget, resume continues); palettes come only from observable page evidence; tech labels are inferences with triggers shown; static-html bands have no measured offsets.


Prompt for AI agents (start here)

✨ The fastest way to a shippable page. Two passes, one confirmation.

Pass 1 — agent calls:
aura_prompt { "goal": "dark SaaS pricing" }

Server returns: draft_prompt + 6 questions + next_call.
Agent MUST ask you (theme / sections / system / assets / macrostructure / vibe) via native question UI.
Pass 2 — agent calls with your answers:
aura_prompt { "goal": "dark SaaS pricing",
  "answers": { "theme": "dark", "sections": ["hero","pricing"],
    "system": "auto-pick", "assets": "no" } }

Server returns: final_prompt (copy-paste) + plan order + suggested calls.

Copy-paste starter you can say to any agent:

Check aura_status. Then call aura_prompt with my goal, ask me its 6 questions,
then call it again with my answers and follow its suggested_calls.
My goal: dark cinematic portfolio.

Rules baked in: Tailwind container mx-auto px-4, hex 50-900, 768px hamburger, hover scale 1.05 + shadow, Inter 40-60/28-36/14-16, motion <500ms + prefers-reduced-motion, tokens first then markup. Sources: tips-for-prompting, introduction, design-systems.


How I use it

New landing page, dark cinematic portfolio, SaaS pricing section. The shape is the same each time:

  1. aura_prompt with the goal sentence. Answer its 6 questions. Copy the final prompt.

  2. aura_recommend with the goal sentence. Skim the starter kit links.

  3. aura_get_component on two or three finalists. Read the markup, not just the description.

  4. aura_install_component on the winner. Follow the file map.

  5. aura_use_design_system once, before any markup, so tokens land first.

Prompts that work well are in SETUP.md.


Free only, no login

There is no auth in this server. No OAuth, no API key to paste, no account to create. Reads go to the same public catalogue endpoints the Aura website uses. The premium flag stays on every component row, free-only is the default, and Pro rows are refused everywhere — search filters them, and direct ID lookup, install, bundle, and scaffold refuse them (-32602, fail-closed) instead of guessing. If Aura rotates its public key, set AURA_SUPABASE_ANON_KEY and rebuild; nothing else changes.

Optional tuning (all with defaults):

Var

Default

Meaning

AURA_TIMEOUT_MS

12000

Upstream timeout

AURA_RETRIES

2

Retry budget for 429/5xx with jittered backoff

AURA_CACHE_TTL_MS

60000

LRU cache TTL (0 disables)

AURA_DEFAULT_LIMIT / AURA_MAX_LIMIT

10 / 25

Pagination defaults and ceiling

AURA_CODE_CHARS / AURA_CONTENT_CHARS

12000

Chunk sizes for code/content paging

AURA_FREE_ONLY_DEFAULT

true

Keep free-only default


Develop

npm test      # hermetic tests, no network (275 tests)
npm run verify  # 45 live checks against the real catalogue + award validation
npm run build   # rebundle src/*.mjs into dist/server.js

Layout: src/config.mjs holds env parsing. src/http.mjs hides timeouts and retries behind an injectable fetch. src/catalog.mjs is the deep module (queries, cache, shaping, counts). src/icons.mjs fetches free icons (FA + pixel lib) with its own 24h cache. src/guide.mjs turns rows into install plans plus pure mappers (buildPrompt, combineScaffold, buildInspo*, buildDocs* — same seam for tools + tests + docs). src/awards.mjs parses/ranks Awwwards references (pure) plus a polite fetch client. src/awards-index.mjs persists the award corpus (atomic JSON, locks, checkpoints, winners, watches). src/browser-cdp.mjs drives optional Chrome over native CDP (screenshots, band scans, motion) with honest unavailability. src/tools.mjs validates args. src/protocol.mjs speaks JSON-RPC. src/server.mjs only wires stdio.


Share

Star it if it saved you a tab or twenty: github.com/swastiksingh-dev/Aura-Components-MCP. Reposits welcome — see DIRECTION.md for where this is headed.

Star History Chart

built a free design MCP on aura.build: 2495 components, 192 skills, 30k assets, 35 tools. no login. resources are @MengTo's

npx -y aura-components-mcp@latest
https://github.com/swastiksingh-dev/Aura-Components-MCP

Credits

Server code by swastiksingh-dev, MIT. Free to fork, use, and sell with. If you fork this, keep the credit line and point people at the original repo.

Contributors

Source

Used for

License / terms

Aura (Meng To / DesignCode)

The entire catalogue: components, skills, assets, design systems

Belongs to Aura and credited authors; assets all-rights-reserved per Aura Terms §4

Official Aura MCP

Comparison only (account projects/publishing over OAuth — the half this server skips)

Aura's terms

Inspo design-reference MCP

Real-site evidence behind aura_inspo_* (recommend, find_components, get_reference_jsx, compare)

Reference JSX belongs to its authors — adapt tokens, don't ship verbatim

Context7

Fresh library docs behind aura_docs_* (trusted: Tailwind CSS, GSAP, Lenis, Iconify)

Docs belong to their projects

Font Awesome Free

Icons via aura_icon (served from jsDelivr + cdnjs)

CC BY 4.0 — credit on every hit, enforced by the tool

HackerNoon Pixel Icon Library (site)

Pixel icons via aura_icon (served from jsDelivr)

Icons CC BY 4.0, other files MIT — credit on every hit, enforced by the tool

Awwwards

Award-website references behind aura_award_* (metadata + Awwwards-CDN media URLs, never mirrored)

Screenshots/designs belong to Awwwards + credited creators — personal-inspiration use at ~1 req/s, honor robots.txt; see THIRD_PARTY_NOTICES.md

awwwards-mcp by INSANE0777

Behavioral inspiration for the six award tools (site discovery, design DNA, elements, briefs, indexing, capture, motion, winners, watches) — reimplemented natively, zero code copied

MIT © 2026 Afjal — permission notice preserved in THIRD_PARTY_NOTICES.md

Upstream libraries (Tailwind CSS, GSAP, Lenis, Iconify)

The code this server's guides point at

Each project's own license — check before shipping


License

MIT — see LICENSE.

Available Tools

35 tools
aura_award_indexB

Local award corpus: status (counts/freshness/lock), rebuild/resume/refresh (bounded polite crawl), new_winners (baseline then delta), watch_add/list/remove (studio/tag/url watches). Never auto-crawls on startup.

Inputs: action (required), kind, award, pattern, +1 more.

aura_award_inspectB

Inspect award refs: site_details (palette/tech/awards/evidence), site_elements (section visuals with media URLs), element (one gallery record), compare_sites (2-3 refs: shared vs gaps, missing marked unknown).

Inputs: mode (required), slug, slugs, elementSlug.

aura_bulk_fetchA

Bulk-fetch 2-8 details in one call for components, design_systems, or assets. Alias-friendly name for aura_bundle. Per-item errors never fail the batch.

Inputs: ids (required), kind (required).

aura_bundleB

Bulk-fetch 2-8 component details in one call (ids or slugs). Per-item errors never fail the batch.

Inputs: ids, slugs.

aura_categoriesA

The 13 component categories with live free counts. Pick one, then search within it.

Inputs: no named parameters.

aura_design_blueprintC

Evidence-backed build plan from 2-5 award refs + goal: Markdown brief + versioned JSON, mapped to real free Aura components/systems (free-only enforced, failures marked unavailable). Never copies third-party source.

Inputs: goal (required), references (required), responsive, technology, +3 more.

aura_docs_checkB

Check pasted code against fresh docs (pure). library + code_excerpt (under 20000 chars, excerpts not full pages) + optional docs_note -> verdict current|needs-update, look-preserving fixes, version-pin needs_check. Fresh API wins, look stays identical.

Inputs: library (required), code_excerpt (required), docs_note.

aura_docs_planB

Plan a fresh-docs lookup (pure, no fetch). library + need -> Context7 resolve/query plan (trusted: tailwind, gsap, lenis, iconify) + apply steps for Aura files + docs stamp slot. The agent calls Context7, Aura shapes the rest.

Inputs: need (required), library (required).

aura_get_componentA

Full free component detail with HTML/Tailwind source, preview image, page URL. Numeric id or slug. Chunk param pages large code (see code_info). full:true returns full code (ignores chunk).

Inputs: id, full, slug, chunk.

aura_get_design_systemB

Full free DESIGN.md content plus preview_html, tokens, and page URL. System id or slug. Chunk param pages large preview_html (see preview_info).

Inputs: id, slug, chunk.

aura_get_skillC

Full free agent-skill content (SKILL.md body) plus source_url and page URL. Skill id. Chunk param pages large bodies (see content_info).

Inputs: id (required), chunk.

aura_iconA

Fetch one free icon as ready-to-paste SVG (live). name in kebab-case (bolt, arrow-down). source auto|fontawesome|pixel, style auto|solid|regular|brands. Auto-fallback across styles then libraries; every hit carries license + credit (CC BY 4.0, credit always) plus a font/CSS alternative. 24h icon cache.

Inputs: name (required), style, source.

aura_inspo_compareC

Triangulate a house style from 2-4 slim Inspo sites (pure). evidence[] of {slug, palette?, fonts?, macrostructure?, mode?} — slim keys only, no autopsy. Returns inspo.compare call + closest Aura system/component queries.

Inputs: evidence (required).

aura_inspo_componentsC

Map an Inspo component filter onto Aura search (pure). evidence.type hero|pricing|features|cta|nav|footer|testimonial|logo-cloud|faq|stat plus style/macrostructure/mode. Returns inspo.find_components call + aura_search + scaffold_hint (tokens first, weight s|m).

Inputs: evidence (required), brief.

aura_inspo_recommendA

Map a brief onto Inspo + Aura calls (pure, no fetch). Strict macro gate: without macrostructure returns need_macro:true — run inspo.recommend first, never scaffold generic. With macro returns inspo_call + 3 aura_calls + palette priority (brand > Inspo > Aura).

Inputs: brief (required), mode, vibe, brand_hex, +1 more.

aura_inspo_referenceB

Adapt one Inspo reference JSX onto one Aura component (pure). evidence {type, id} + aura_component {id|slug}. Returns inspo.get_reference_jsx + aura_get_component calls plus scoped adapt_plan (token swaps, css scope, files).

Inputs: evidence (required), aura_component (required).

aura_inspo_systemB

Resolve Inspo tokens vs Aura DESIGN system (pure). inspo_tokens {palette, fonts, background?, text?, brand_hex?} + aura_system slug. Returns option A (Inspo) vs B (Aura fetch) vs C (blend) plus one system_choice question payload (A-inspo|B-aura|C-blend) and contrast warning. Tokens first.

Inputs: aura_system (required), inspo_tokens (required).

aura_install_assetC

Legal drop-in plan for an asset: direct download URL, preview URL, license (all-rights-reserved per Aura Terms §4 — check page before commercial use), suggested file path.

Inputs: id (required).

aura_install_componentB

Paste-ready setup for a free component: dependency list, setup steps, file map, fonts. Goes beyond the official Aura MCP, which only reads project source.

Inputs: id, slug.

aura_install_skillC

Save-and-load plan for a free skill: where to put SKILL.md per client plus upstream source.

Inputs: id (required).

aura_motion_studyB

Evidence-backed motion guidance. Modes: inspect (live runtime scan: libs/easing/scroll), record (bounded frame study; .webm only when ffmpeg present), search (stored studies by library/easing/scroll). Never claims unobserved library state.

Inputs: mode (required), url, easing, frames, +5 more.

aura_promptA

Need-based design prompt in 2 passes: Pass 1 (goal only) returns draft + 6 questions (theme, sections, system, assets, macrostructure, vibe — mode folds into theme); ask the user via native question UI, then Pass 2 (goal + answers) returns copy-paste final_prompt + plan order + suggested calls starting at inspo.recommend. Always confirm with user before scaffolding.

Inputs: goal (required), theme, system, answers, +1 more.

aura_recommendB

Starter kit for a goal: top 3 free hits per surface with page URLs and rationale. Slim rows + commercial-use note. Empty components auto-retry once via suggested_queries, then editorial picks.

Inputs: goal (required), slim, theme, min_views, +1 more.

aura_scaffold_pageB

One ordered page build: DESIGN.md tokens.css + system preview + component markup in dependency order, combined deps + files[]. Merges install_* + use_* in a single turn. components accepts string|number|{id}|{slug} refs, e.g. ["732","1364"] or [{id:"732"}] or [{slug:"..."}]; full:true returns full bodies.

Inputs: goal (required), full, system, components.

aura_search_allA

One call across components, skills, assets, and design systems in parallel. Free only. Slim rows by default (excerpts, no blobs). Each surface runs its own fallback pipeline, so mixed outcomes are normal (skills may fall back while components go honestly empty) — read isFallback/no_match per surface, not once globally.

Inputs: slim, limit, query, theme, +3 more.

aura_search_assetsC

Search free Aura assets (images and video). Keywords, media_type image or video. Slim rows by default. exact_match:true skips fallbacks.

Inputs: slim, sort, limit, query, +4 more.

aura_search_componentsA

Search free Aura UI components (2,495). Text over title and description, optional category tag, theme dark|light filter, negative_theme exclusion, min_views floor, sorts. Slim rows by default (excerpt + fetch pointer, no full code; slim:false for full). exact_match:true skips fallbacks (honest empty).

Inputs: slim, sort, limit, query, +6 more.

aura_search_design_systemsA

Search free Aura DESIGN.md systems (725). Metadata + snippets, slim by default; use aura_get_design_system for content. exact_match:true skips fallbacks.

Inputs: slim, sort, limit, query, +2 more.

aura_search_skillsA

Search free Aura agent skills (192). Metadata only, slim by default; use aura_get_skill for the full SKILL.md content. exact_match:true skips fallbacks.

Inputs: slim, sort, limit, query, +2 more.

aura_statusA

Catalogue health plus free counts (components, skills, assets, design systems). Free only, no login. Start here.

Inputs: no named parameters.

aura_use_design_systemC

Apply a free DESIGN.md system: token starter CSS plus copy order (tokens first, then markup).

Inputs: id, slug.

aura_versionA

Running vs npm-published version (live, hourly cache). Returns update_available + notice telling the agent to prefer latest (npm i -g aura-components-mcp@latest) when behind. Never fails — offline registries return latest:null. aura_status already carries this verdict on every call.

Inputs: no named parameters.

aura_visual_auditB

Capture/analyze a reference or your build. Modes: capture (real browser screenshot), structure (section band map: live-dom or static-html), compare (reference vs build mismatches). Needs Chrome for live capture; honest BROWSER_UNAVAILABLE otherwise. Loopback only with allowLocal:true.

Inputs: mode (required), url, buildUrl, settleMs, +4 more.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    B3/5.0

    Scored across 35 tools

    Disambiguation3/5

    Four family clusters (aura_inspo_*, aura_award_*, aura_search/get pairs) are mostly distinct, and descriptions do a lot of boundary-setting. However aura_bulk_fetch is an explicit alias of aura_bundle, and install vs scaffold vs use_design_system overlap enough that an agent could misselect. With 35 tools, several boundaries (recommend vs search_all, install_component vs scaffold_page) require careful reading.

    Naming Consistency4/5

    Every tool uses the aura_ prefix in lowercase snake_case, with clear verb_noun names (aura_search_components, aura_get_component, aura_install_skill). Minor deviations are noun-only names (aura_status, aura_categories, aura_icon, aura_version) and namespace prefixes (inspo/award/docs), but the convention stays readable and predictable overall.

    Tool Count2/5

    35 tools is well above the 25+ threshold, and the surface sprawls across roughly a dozen sub-domains (components, skills, assets, systems, awards, motion, docs, inspo, audit). The redundant aura_bundle/aura_bulk_fetch alias confirms some tools do not each earn a distinct place.

    Completeness4/5

    Coverage is broad and coherent for a read-only catalog: search+get+install for components/skills/systems/assets, plus aggregators (search_all, recommend, related, bundle), awards, motion, docs, and visual audit. Minor gaps like no dedicated asset detail/detail-get tool exist, but core discovery-to-scaffold workflows are complete.

    Maintenance

    ActivityMaintained
    ResponsivenessOutdated data

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      A
      quality
      C
      maintenance
      Enables MCP-capable agents to browse a public design-system catalog, retrieve design continuation bundles and sequential prompt packs, import externally designed artifacts, and resume mid-flow external design runs.
      8
      75 npm
      MIT
    • A
      license
      Not graded
      quality
      C
      maintenance
      Enables MCP-aware clients to query and use the you-design catalog of design systems, skills, and plugins at runtime via stdio.
      Apache 2.0
    • A
      license
      Not graded
      quality
      B
      maintenance
      Enables MCP clients to search skills.sh, view leaderboards, inspect skill details, retrieve third-party security audits, and browse official curated skills, with install commands for each result.
      41 npm
      MIT
    • A
      license
      Not graded
      quality
      B
      maintenance
      Enables MCP clients such as Claude, Claude Code and Cursor to keyword-search a directory of over 1,000 hand-picked productivity tools, apps, AI tools, courses and books, filtering by category, type, pricing and result limit, and to fetch a single listing by slug, page URL or exact name and list all categories with counts. It runs read-only over stdio with no API key, dependencies or tracking, making only a plain GET of the public catalogue.
      MIT