Design Atlas MCP
Provides search and retrieval of Bootstrap Icons SVGs, including icon names, copyable SVG content, and attribution/license information.
Provides official links and client instructions for using Linear as a design-reference source through the reference guide tool.
Provides search and retrieval of Lucide SVG icons, including copyable SVG content and license attribution.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@Design Atlas MCP find an editorial reference for my ferry brand and give me its HTML, CSS and tokens"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Design Atlas MCP
Design Atlas is an open library of working interface studies for designers and coding agents. Search a reference, try the interaction, inspect its HTML and CSS, and turn it into a kit for your brand.
Try the guided demo · Browse 36 studies · Open the brand studio · Get the MCP release
A quiet product introduction | Working issue queue | An editorial motion study |
|
|
|
Use it today
To try | Open | What you can do |
A working product pattern | Choose a study, interact with it, inspect the source and adapt it to a brand. | |
A microinteraction | Hold a berth, release it, then inspect the seven motion frames and the code behind the transition. | |
Your own visual identity | Paste existing HTML and CSS or enter a brief. Preview the kit, check color contrast and download the HTML, stylesheet and tokens. |
Watch the 14-second recording to see the reservation study before and after a brand change. Download the Harbor Press starter kit, unzip it, and open index.html to explore the generated site, design tokens and reusable components.
Related MCP server: Niblet
Connect a coding agent
Install the MCP once. Download the release tarball into the current directory, then install that local file with Node.js 22.16 or newer. This works with npm 12’s default remote-package policy.
curl -fLO https://github.com/Tradelord223/design-atlas-mcp/releases/download/v1.0.1/design-atlas-mcp-1.0.1.tgz
npm install -g ./design-atlas-mcp-1.0.1.tgz
design-atlas-mcp infoYou can also download the package in your browser. The release includes SHA256SUMS.
In Codex, register the installed executable and open a fresh chat.
codex mcp add design-atlas -- design-atlas-mcpIn Claude Code, register the same executable.
claude mcp add --transport stdio design-atlas -- design-atlas-mcpIn Cursor or Claude Desktop, add this entry to the client’s MCP configuration.
{
"mcpServers": {
"design-atlas": {
"command": "design-atlas-mcp",
"args": []
}
}
}For a GUI client with a restricted PATH, use node as the command and the absolute path to design-atlas-mcp/src/cli.js inside the directory printed by npm root -g as its only argument. The first installation downloads dependencies. All core MCP tools operate offline afterward.
Try a design brief
Harbor Press publishes pocket field guides for ferry riders. Our ink color is #345d4e. Find two relevant editorial references and show me what each contributes. Build a responsive sample using our name, color and audience. Give me the tokens and HTML, and flag anything that still needs editorial or accessibility review.
For an existing product, paste its HTML and CSS into the workbench or use import_brand_identity. Review the identified colors, type, corner radii and the declarations they came from. Give the agent a specific component and its source. Ask it to preserve the brand typography and the component’s important states, compare two Atlas treatments, and apply the selected change. A screenshot or source link accompanies each study. Atlas previews show original implementations. A source link identifies the product or design-system example behind the study.
Included in v1.0.1
36 studies covering landing pages, notes, command menus, issue triage, forms, commerce, editorial layouts, financial dashboards, mobile onboarding, studios and motion. Each includes source attribution, an implementation and a versioned evidence bundle.
Nine design packs with design decisions and tokens. Choose Quiet Product, Precise Workspace, Public Service, Material Commerce, Editorial Reference, Financial Clarity, Daily Companion, Independent Studio or Motion Printshop. The gallery shows the studies in each direction.
Eight product flows with ordered steps, a pattern for each step, its goal and a failure or recovery consideration.
4,805 offline SVGs across Lucide (2,403), Bootstrap Icons (2,078) and Heroicons 24px outline (324). Search by name and get an actual SVG with its license.
A brand workbench and source audit. Generate responsive HTML, CSS, tokens and reusable components. Review copy, labels, document semantics, keyboard focus and motion rules. Palette checks report measured ratios. Static audits identify what still needs a rendered review.
Remix tools. Export a complete specimen, isolate a named element, preview three changes side by side, or assemble selected sections and attribute each choice. A portable headline entrance can be attached to the first section in a composition.
MCP tools
Task | Tools | What your agent receives |
Search and inspect |
| Ranked candidates, stable IDs and source links. Inspection returns an inline PNG or seven ordered motion frames, plus a versioned manifest. |
Read evidence |
| Bounded UTF-8 file transfers with SHA-256 verification, or numbered source lines with character continuation. |
Use the Vault |
| Implementation prompt and working source, documentation or extras drawn from the same Atlas studies. |
Explore systems |
| Pack tokens, design principles, linked specimens and complete product journeys. |
Use icons |
| Installed families, paginated names, copyable SVG and attribution. |
Design for a brand |
| Evidence from supplied brand markup and CSS, brand-specific output, and issues with concrete fixes. |
Adapt a reference |
| A standalone rebuild, isolated detail, variations or composition proof from selected references. |
Open a source library |
| Official links and client instructions for Details, Mobbin, Granola, Linear and six other design sources. A pasted Details link receives the matching official tool handoff. |
The server also exposes atlas://library/index as an MCP resource and design_from_brief as a reusable prompt.
Search and inspect
Send the user's words as
querywithprojectContext. Optionally passprojectBrief,category,platformorpackId. Search uses local lexical and synonym ranking. Each gallery contains at most ten candidates.Present the candidate links. To see another set or refine the terms, carry over
nextSearchArguments, including its cumulativeexcludeIds.Use
get_inspirationwith the selected Atlas ID or gallery URL. Its default response contains the visual evidence. SetincludeImages: falsefor the metadata alone.Read
instructions.mdbefore the files it cites. Supply the returnedversionon every subsequent read. For numbered source, follownextOffsetin the same line range before advancing tonextStartLine. Transfer mode uses byte offsets and base64 chunks. Text mode counts Unicode code points.Use
get_vault_resourcefor the code orremix_inspirationsfor a reviewable adaptation. Interactive demos identify their integration points for real data and service calls.
Run from source
git clone https://github.com/Tradelord223/design-atlas-mcp
cd design-atlas-mcp
npm ci
node src/cli.js info
node src/cli.js kit examples/harbor-press.json ./harbor-kitOpen harbor-kit/index.html for the brand preview. The new directory also contains brand.css, tokens.json, individual components and design-notes.json. The exporter requires a fresh output directory.
npm start # MCP over stdio
node src/cli.js http --port 8787 # Streamable HTTP at http://127.0.0.1:8787/mcp
npm test # protocol, evidence, search, brand, icon and browser tests
npm run gallery # static site in dist/siteBrowser checks use installed Chrome on macOS or CHROME_PATH elsewhere. The GitHub workflow supplies Chrome and verifies the browser tests before deploying Pages. To regenerate a specimen’s visual evidence, run CHROME_PATH=/path/to/chrome npm run render -- opening-imprint. With no IDs, the renderer captures the entire authored library.
For shared HTTP hosting, put the Node service behind HTTPS, set ATLAS_HOST=0.0.0.0 and set ATLAS_PUBLIC_ORIGIN to the external origin. The HTTP server checks Host and Origin, accepts requests at /mcp, and holds no accounts or user content. A private deployment can enforce its own access controls at the reverse proxy. GitHub Pages serves the static gallery and workbench. A Gradio deployment adapter hosts the complete MCP at /mcp and adds optional screenshot similarity search on ZeroGPU. Hugging Face deployment files
Add your own design packs
From a checkout, run npm run add-pack -- /path/to/pack.json. The importer checks the pack’s IDs, rights declarations, references and schema before updating the local catalog. Render the new studies, then rebuild the gallery. The same studies become available to MCP search, inspection, the Vault and the brand workbench. Pack authoring guide and example
Provenance and licensing
The six core tool names and design workflows were independently reconstructed from Details’ public MCP guide, public client artifacts and unauthenticated protocol discovery. The repository contains the observed evidence, hashes and limits of that comparison. Details’ authenticated schemas, private ranking, captures and Vault corpus are outside the verified comparison. Mobbin and Details users with access to their hosted collections can open their official connectors through get_reference_guide. Reconstruction record · Details MCP · Mobbin MCP
Original implementation, studies and previews are MIT licensed. Icons retain their upstream licenses and attribution. Product names and linked design references belong to their respective owners. Atlas license · Lucide · Bootstrap Icons · Heroicons
Available Tools
18 toolsaudit_designBRead-onlyIdempotent
Inspect supplied HTML, CSS and copy for actionable issues. Returns evidence, fixes and scope.
Inputs: css, copy, html.
create_brand_kitBRead-onlyIdempotent
Build a brand-specific responsive kit. Default returns the source manifest, tokens and contrast checks. Repeat the same brief with output=file and one manifest path to receive its content. Full returns the whole kit.
Inputs: name (required), description (required), path, limit, +13 more.
get_design_packBRead-onlyIdempotent
Read a pack’s appearance tokens, principles and reference IDs.
Inputs: id (required).
get_flowBRead-onlyIdempotent
Get ordered steps, reusable pattern IDs and recovery considerations for a product journey.
Inputs: id (required).
get_iconBRead-onlyIdempotent
Get the complete inline SVG, its upstream version, license text and accessible usage.
Inputs: name (required), pack (required).
get_inspirationARead-onlyIdempotent
Inspect a selected reference. Returns an actual PNG or seven ordered frames, plus a versioned source manifest. Set includeImages to false for metadata only.
Inputs: id, slug, includeImages.
get_reference_guideCRead-onlyIdempotent
Open the official provider workflow for an upstream reference. Returns vetted links and native MCP handoff instructions for Details, Mobbin and design-system sources.
Inputs: provider (required), url, topic.
get_vault_resourceCRead-onlyIdempotent
Get source and an implementation prompt, supporting documentation, or extras for a Vault ID.
Inputs: id, slug, content.
import_brand_identityARead-onlyIdempotent
Extract declared brand colors, font families, radius, name and description from supplied HTML or CSS. Returns the evidence, unknowns and a brief for create_brand_kit. Runs locally.
Inputs: css, html, name, overrides, +1 more.
list_design_packsARead-onlyIdempotent
List visual systems with their principles and number of linked patterns.
Inputs: no named parameters.
list_flowsCRead-onlyIdempotent
Find multi-screen journeys with steps, goals and failure states.
Inputs: no named parameters.
list_icon_packsARead-onlyIdempotent
List installed SVG families with versions, counts and attribution.
Inputs: no named parameters.
read_inspiration_bundleBRead-onlyIdempotent
Read a versioned file. Text offsets count Unicode code points. In transfer mode, concatenate decoded base64 byte chunks and verify the complete SHA-256.
Inputs: id, mode, path, slug, +3 more.
read_inspiration_sourceCRead-onlyIdempotent
Read numbered source lines. Continue with nextOffset in the current range, then nextStartLine at offset zero.
Inputs: id, path, slug, limit, +4 more.
remix_inspirationsCRead-onlyIdempotent
Build from a reference, extract a named detail, generate three side-by-side variations, or combine selected sections with attributable motion.
Inputs: ids (required), mode (required), detail, headlineMotion.
search_iconsBRead-onlyIdempotent
Search local SVG names and families. Use the exact returned pack and name with get_icon.
Inputs: pack, limit, query, offset.
search_inspirationsBRead-onlyIdempotent
Find authored reference studies. Present the returned candidates with source links, then inspect the selected ID. Continue with nextSearchArguments.
Inputs: limit, query, offset, packId, +5 more.
search_vaultCRead-onlyIdempotent
Find implementation-ready snippets and interaction studies by task, category, pack and platform.
Inputs: limit, query, offset, packId, +5 more.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 18 tools
The set is organized into clearly separated domains (icons, brand, audit, inspirations, vault, flows, packs), and each tool names a distinct resource plus action. The main soft spots are the parallel retrieval triples (search_inspirations/get_inspiration vs search_vault/get_vault_resource) and the read_inspiration_bundle/read_inspiration_source pair, which an agent could pick wrongly, plus remix_inspirations which bundles four different behaviors into one tool.
All names are snake_case verb_noun and follow the list_/search_/get_/read_/create_ pattern per resource domain, which is highly predictable. Minor deviation: 'get' and 'read' are both used for retrieval-style tools (get_inspiration vs read_inspiration_source), blurring the verb semantics slightly.
Eighteen tools is slightly above the ideal range but justified by the breadth of the domain: icons, brand kits, audits, inspirations, vault snippets, reference guides, packs and flows. Each tool corresponds to a real operation, with only marginal redundancy in the inspiration reading tools.
Coverage is strong across every sub-domain: listing, searching and fetching for icons, inspirations, vault resources, designs packs and flows, plus creation and auditing workflows (import_brand_identity, create_brand_kit, audit_design, remix_inspirations) and chunked file reads for large bundles. No obvious dead ends for the read/reference-oriented purpose of this server.
Maintenance
Related MCP Connectors
A maintained design library your coding agent can call.
Access and maintain design system docs, tokens, components, skills, and contexts across any project.
Live React design-system APIs, patterns, and code validation so AI agents build real UI, not slop.
Give your agent a real design system: tokens, measured WCAG contrast, and rules to follow.
Related MCP Servers
- AlicenseAqualityBmaintenanceIt lets AI coding agents look up design tokens, component specifications, and prohibition rules on demand, then validate generated HTML or JSX with the same lint logic used by CI and editor hooks so violations are caught and corrected immediately. It also exposes the design constitution and rule sets as resources for reference during generation.6206 npm202MIT
- FlicenseNot gradedqualityAmaintenanceGives coding agents concrete, real product screens as inline references for unresolved layout, state, or interaction questions, plus licensed fonts and icons and the recorded design system behind a chosen web screen. Also bundles a token-free design workflow, diagnostics, and command playbooks that keep UI work grounded in the agent's own product.-
- AlicenseNot gradedqualityAmaintenanceEnables coding agents to ground UI work in real product screens by searching a catalogue for concrete screen references and design materials such as fonts, icons, and animated icons, then pulling the colors, typography, components, and provenance behind a chosen web screen. It also serves bundled skill documents, command playbooks, and connection diagnostics so an agent can settle a screen's states and acceptance criteria before building and verify the rendered result afterward.MIT
- AlicenseNot gradedqualityAmaintenanceEnables coding agents to pull real product screens as inline image references, find licensed fonts and icons, and retrieve recorded design-system details for web screens so UI work is grounded in actual interfaces rather than generic templates.MIT


