Brand Kit MCP
Allows Figma Make and the Figma agent to retrieve brand tokens, visual rules, layout suggestions, examples, and design checks so generated designs stay on-brand.
Allows Replit Agent to access brand overview, tokens, rules, format recipes, layouts, examples, and design/copy checks while building projects so its output follows the brand kit.
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., "@Brand Kit MCPGet my brand’s main colours, fonts and top rules for a landing page"
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.
Brand Kit MCP
An open-source MCP server that gives any AI tool a brand to follow. It serves a Brand Kit, a small folder of plain files describing how a brand looks, sounds and adapts to different formats, and checks AI output against it.
Connect it to Claude, ChatGPT, Figma Make, Replit or Cursor, and what they make comes out on-brand.
All tools are plain code. The server never calls an AI model, so it runs free on Cloudflare's free plan.
Tools
Tool | What it does |
| Identity, audience, values, facts and guardrails. The AI calls this first. |
| Exact colours with roles, fonts, spacing, radius (DTCG format). |
| Visual or voice rules, each marked must or should, with reasons. |
| How the brand translates to LinkedIn posts, carousels and slides. |
| Three different on-brand layouts for a piece of content, so designs vary. |
| On-brand work, labelled as examples rather than templates. |
| Checks HTML, CSS or SVG: palette, near-miss colours, fonts, corners, decoration, contrast. |
| Checks text: spelling variant, em dashes, words to avoid, LinkedIn rules. |
| Reads a public site's CSS: exact colours, variables, fonts, sizes, spacing, ranked by use. |
Related MCP server: OnBrand by SlideSpeak
Run it on your computer
You need Node.js 20 or newer.
npm install
npm run devThe server runs at http://localhost:8787/mcp. In a second terminal, run the smoke test, which connects like a real AI tool and calls every tool:
npm testTo try it inside Claude Code while it runs locally:
claude mcp add --transport http brand-kit http://localhost:8787/mcpPut it online (free)
Create a free Cloudflare account.
In this folder, run:
npx wrangler login npm run deployWrangler prints your server's address, like
https://brand-kit-mcp.<your-subdomain>.workers.dev. Your connector URL is that address plus/mcp.
The free plan allows 100,000 requests a day.
Use your own domain
Once your domain's DNS is managed by Cloudflare, uncomment the routes line in wrangler.jsonc and deploy again. For example, brandkit.soraiasoares.com/mcp.
Connect it to AI tools
Add the connector URL as a custom MCP connector in each tool. Menus change often, so check each tool's help if these names have moved.
Claude Code:
claude mcp add --transport http brand-kit https://<your-address>/mcpClaude (web and desktop): custom connectors in Settings. Some users have reported that claude.ai asks public servers without login to set one up; if that happens, use Claude Code or Claude Desktop meanwhile.
ChatGPT: turn on Developer Mode, then create a custom connector with the URL. Needs a paid plan.
Replit: Integrations, then MCP Servers for Replit Agent, then Add MCP server.
Figma Make and the Figma agent: custom connectors. Needs a paid plan and a public HTTPS address (not localhost).
Cursor and VS Code: add the URL as a remote MCP server.
Add your own kit
Copy
kits/soraia-soarestokits/<your-id>and replace the files with your brand's:brand.md: identity, audience, values, facts and guardrailstokens.json: colours, fonts, spacing, radius (DTCG format)visual-rules.mdandvoice.md: rules as- id (must|should): one sentenceformats.md: motifs, constants, variables, format recipes, archetypesexamples/README.md: on-brand work
Import the files in
src/kits.tsand add an entry toKITS.Your kit is served at
/kits/<your-id>/mcp. SetDEFAULT_KITinwrangler.jsoncto serve it at/mcp.
The checks read your kit: check_copy only applies a rule like no-em-dash or british-english if your voice.md declares it.
Privacy: anyone with the URL can read the kit. Keep confidential material out of it, or deploy your own private copy.
Project structure
kits/<id>/ the Brand Kit files
src/index.ts routes: /, /mcp, /kits/<id>/mcp
src/server.ts the MCP tools
src/checks.ts check_design and check_copy
src/styles.ts read_website_styles
src/layouts.ts suggest_layouts
test/smoke.mjs end-to-end test with a real MCP clientLicence
MIT
This server cannot be deployed
Maintenance
Related MCP Connectors
Serves your design system and coding standards to coding agents, so they stop guessing.
Your brand's Styles, logos and asset library in your AI app, for on-brand assets that match.
Brand-safe MCP for AI agents to create editable, on-brand graphics and automate variants.
Brivvy MCP (OAuth): on-brand AI content, voice rules, templates, glossary.
Related MCP Servers
- AlicenseNot gradedqualityDmaintenanceExtracts brand identity (colors, typography, spacing) from any website and generates HTML/CSS replicas or applies branding to templates.1MIT

OnBrand by SlideSpeakofficial
AlicenseNot gradedqualityCmaintenanceAn MCP server that feeds AI agents your brand's real logos, colors, fonts, and approved slide layouts, so every deck and document comes out on brand the first time.7MIT
user-majico-mcpofficial
AlicenseNot gradedqualityBmaintenanceMCP server for Majico.xyz that enables coding agents to read (and limited write) brand guidelines, design tokens, studio canvas, and export manifests.18 npmMIT- AlicenseNot gradedqualityBmaintenanceMCP server that exposes brand identity guidelines (visual look and voice) as markdown, enabling LLMs to produce on-brand content.19 npmMIT