Set up your agent

Install once, and your agent just knows your brand.

BrandSweets ships as a Claude Code plugin: a read-only MCP server scoped to your workspace, plus four design skills. Your agent reads your real brand system — colors, type, logos, voice, usage rules — and designs against it. Three steps, about two minutes.

Get an API key

In the app, open Settings → Workspace API keys → New key. Choose a write key (bsk_w_…) so your agent can edit brands and run the creation and logo loops — a read key (bsk_r_…) grounds in your brands read-only. Copy the token — you'll paste it once; rotate or revoke it anytime from the same screen.

You'll also need your workspace slug — the {workspace} segment in your MCP URL.

Add the MCP config

Point your agent at the workspace MCP endpoint and pass the token as a bearer header. One URL exposes every brand in the workspace; the agent calls list_brands to enumerate them.

MCP server config

{
  "mcpServers": {
    "brandsweets": {
      "type": "http",
      "url": "https://mcp.brandsweets.com/w/<workspace>/mcp",
      "headers": { "Authorization": "Bearer <workspace-api-token>" }
    }
  }
}

Substitute your <workspace> slug and <workspace-api-token>. If you install the plugin (next step), this file is generated for you from the values you enter.

Install the plugin (or the skills)

The plugin bundles the MCP config and the four skills together. In Claude Code:

Claude Code

# 1. Add the marketplace (published repo, URL, or local checkout)
claude plugin marketplace add ethanplusai/brandsweets

# 2. Install the bundle
claude plugin install brandsweets-skills

# 3. When prompted, enter your workspace slug + API token

Prefer to wire it by hand? Add the MCP config above and drop the skills into your project's .claude/skills/:

Manual

# Copy the skills into your project (or ~/.claude/skills)
cp -R skills/brandsweets-brand-design \
      skills/brandsweets-logo-design \
      skills/brandsweets-on-brand \
      skills/brandsweets-imagery \
      skills/design-taste \
      .claude/skills/

What your agent can do now

Start a session and ask it to use a brand:

Ask

Use my "acme" brand and build a landing-page hero.

The right skill loads automatically and the agent grounds in your brand before it writes a line of code.

Build on-brand

Fetches your brand system first, maps tokens onto Tailwind / shadcn / plain CSS, uses your contrast pairs, and screenshot-reviews the render against the brand. No invented hexes or fonts.

brandsweets-on-brand

Design the brand

Runs the creation loop: proposes genuinely distinct directions, honors locks as hard constraints, consumes typed feedback, and converges to a locked, coherent system. With a write key it can create a brand from nothing (create_brand) and build it up.

brandsweets-brand-design

Design logos

Starts a logo round (start_logo_round), submits original vector marks against a construction grammar, and runs the deterministic render → QA → critique loop (submit_logo_concepts, get_logo_qa, derive_logo_variants) — you lock the winner on the canvas.

brandsweets-logo-design

Generate imagery

On-brand image generation (Pro + your own provider key). You supply the subject; the brand's reference-image style is applied automatically. Offers candidates, never "pixel-perfect."

brandsweets-imagery

Better taste, everywhere

Framework-agnostic frontend craft — reads the brief, rejects generic defaults, keeps motion honest, and audits against evidence. Works even before a brand exists.

design-taste

A read key exposes the read tools only; a write key additionally unlocks brand edits, the creation and logo loops, and Pro imagery — all on the one workspace endpoint. Edits from a write key auto-approve (a new version is minted immediately), since the key is your own trusted agent. Works in Claude Code today; any agent that can register a remote MCP by URL can use the server directly.

Create a brand See pricing