brandsweets

Brand memory for the agent era

Your brand, sweetened for agents.

Extract or design a brand, refine it alongside your agent, then serve it to every future session as structured, agent-readable data. Connect once — everything your agent builds comes out on-brand.

Extract my brandCreate a brand
brandsweets.com/b/stripe
Stripe
stripe.com
live
primary
#635BFF
accent
#00D4FF
ink
#0A2540
surface
#F6F9FC
brand.md
tokens.json
mcp
---
name: Stripe
voice: [precise, calm, technical]
---

# Payments infrastructure
Write like a developer talking to a
peer. Lead with the API. Never hype.
mcpbrandsweets.com/b/stripe/mcpagent-ready
Path 1

I have a brand

Type your domain. We run a real browser over it and extract colors, type, logo and voice into a reviewable draft.

Extract my brand
Path 2

I need a brand

Start from a brief — name, one-liner, audience, vibe — and shape it into a complete brand system on the canvas.

Create a brand

Extraction

Type a domain. Get a structured brand — colors, type, logo and voice, pulled from the real site.

Domain
stripe.comExtract
headless browser · reads the live site
Stripe
stripe.com
draft
#635BFF
#00D4FF
#0A2540
#F6F9FC
Aa Söhneprecisecalmtechnical
Already extractedLinearlinear.appNotionnotion.soVercelvercel.com

The loop

Your agent proposes directions. You react on the canvas. The winner locks — and constrains everything after.

agentpropose 3 directions for Acme Robotics
propose react lock
v1 · Geometricproposed
Aa
Acme Robotics
#4F46E5
v2 · Grotesquelocked
Aa
Acme Robotics
#635BFF
v3 · Editorialproposed
Aa
Acme Robotics
#0EA5E9
v2 lockedbecomes a hard constraint every future session reads first.

One document, three surfaces

Every brand serializes losslessly to the formats agents actually consume — no bespoke SDK, no lock-in.

brand.mdRead

A narrative brief agents ground on before they write a word — voice, story, dos and don'ts, in one markdown file.

tokens.json · brand.cssApply

Role-based color, type, spacing and radius as JSON plus CSS variables. One import and a prototype is on-brand by construction.

mcpConnect

A per-brand MCP endpoint. Agents fetch the right logo, tokens and copy — and later propose changes humans approve.

Connect your agent

Every brand is one MCP endpoint away. Paste it in, and the session grounds itself in the real system.

.mcp.json
{
  "mcpServers": {
    "brandsweets-stripe": {
      "type": "http",
      "url": "https://mcp.brandsweets.com/stripe/mcp"
    }
  }
}
# or: claude mcp add --transport http brandsweets-stripe …/api/mcp/stripe/mcp
Without

“Here’s a blue button.”

generic · guessed
With brand

Söhne 500 · brand primary

Get started
#635BFF · on-brand
Any MCP client

Claude Code, Cursor, your own agent — if it speaks MCP, it reads the brand. Streamable HTTP, no SDK.

Read-only, grounded

Tools return the real tokens, voice and rules; a brand://{slug}/system resource grounds a session in one call.

Every session, current

Re-lock the brand and subscribed sessions get pushed the update. On-brand by construction, not by re-prompting.

Imagery, on-brand

Steer generation with a style and your tokens — bring your own key.

Style
warmeditorialsoft daylightfilm grain
Generated
on-brand
on-brand
geminirecraftbyokPure pass-through — your key, your inference. We never proxy or resell.

In the corpus

All brands

Hundreds of real brands, already addressable — each a page plus machine endpoints an agent can query.

FAQ

Questions agents and their humans ask.

What is an agent-native brand?
An agent-native brand is a brand stored as structured, machine-readable data — role-based color tokens, typography, logo inventory, voice and usage rules — instead of a static PDF. BrandSweets serves every brand as brand.md, tokens.json and brand.css plus a Model Context Protocol (MCP) endpoint, so any AI agent can read the real system and stay on-brand.
How do I make my AI agent use my brand?
Connect the brand's MCP endpoint to your agent. For public read-only access point it at mcp.brandsweets.com/<slug>/mcp; for the full read-write loop use your workspace endpoint mcp.brandsweets.com/w/<workspace>/mcp with a workspace API key as a Bearer token. The agent then fetches your real colors, type, logos and rules before it writes any code or copy.
How do I connect Claude Code to BrandSweets?
Install the plugin: run `claude plugin marketplace add ethanplusai/brandsweets` then `claude plugin install brandsweets-skills`, and enter your workspace slug and API token when prompted. This wires the MCP server and the BrandSweets design skills. You can also add the MCP server by hand with `claude mcp add --transport http`. See /docs/agents for the full walkthrough.
What formats does BrandSweets expose?
Each brand serializes losslessly to brand.md (a narrative brief agents ground on), tokens.json (role-based W3C DTCG design tokens), and brand.css (CSS custom properties). All three are available at stable URLs under /b/<slug>/, alongside a per-brand MCP endpoint.
Is BrandSweets free?
Yes — one brand is free forever at full fidelity, no credit card, including hosted logos, exports and a working MCP endpoint. Pro is $19/mo for unlimited brands, an unthrottled priority MCP endpoint, corpus search, imagery generation, PDF export and CI brand-drift checks, with team seats at $5/mo each.

More for agents: /docs/agents, /llms.txt, and every brand's brand.md.

One brand is free forever — full fidelity, no card, a working MCP endpoint. Pay only for more brands or a team.

See pricing

Ready to sweeten a brand?

Free while we build. Extraction drafts stay private to you.

Have a brandNeed a brand