Brands have always had an interface. It used to be a person: the brand manager you emailed to ask whether you could use the old logo on a banner. Then it was a PDF. Now it’s an API, and that changes who can ask the question.
Every brand in Brandbook has one. It’s not an add-on and there’s no extra setup. Make a key, and your brand is something code and agents can read, check against and update.
What you can ask it
The REST API is small on purpose. Five things cover most of what a tool needs from a brand:
- List the brands in your workspace.
- Get a brand’s whole system: colours, type, logo, voice, imagery and rules, as structured data and as brand.md.
- Check copy, a design or a palette, and get a score with findings back.
- Ask a question, like “which colour goes on Lavender?”, and get a short answer.
- Add a rule or update one section of the brand, logged in the changelog like any other edit.
Here’s a check. Send the draft, get back a score, a verdict and what to fix.
curl -X POST https://brandbook.app/api/v1/check \
-H "Authorization: Bearer $BRANDBOOK_KEY" \
-H "Content-Type: application/json" \
-d '{"brand": "late-club", "text": "Only 2 left!!! Drop 05 is selling fast"}'The answer names each problem and, where it can, the fix: “only 2 left” and “selling fast” break the no-fake-scarcity rule, and the exclamation marks come out. Your code can apply those fixes or show them to a person. Every check is logged on the brand’s Check page, so you can see what your tools have been sending.
For agents: MCP
If you use Claude Code, Claude Desktop, Cursor or VS Code, you don’t need to write any code at all. They speak MCP, the protocol AI tools use to call other tools. Add Brandbook’s server address and your key to their settings, and the agent can list your brands, read a brand’s context, check its own output, ask the brand a question and add a rule.
In practice that means you can ask Claude for three captions for Drop 05, and it will fetch Late Club’s voice rules first, write the drafts, run a check, and only show you the ones that pass. You didn’t paste a style guide. You didn’t remind it twice.
The best brand guidelines are the ones you never have to paste.
One key, every brand
A key belongs to a workspace, not a brand. If you run several brands, one key covers all of them and the agent picks the right one by name. Keys start with bb_live_, are shown once when you make them, and can be revoked any time from Settings › API & connections.
If you don’t write code
You still benefit. The API is what lets the tools your team already uses stay on-brand without anyone asking. And for the tools that don’t speak MCP yet, there are plain files: brand.md, a readable summary you can drop into a ChatGPT or Claude project, and tokens.json, your colours and type in the Design Tokens format that Figma plugins and front-end code read.
It works without AI, too
Checks run on a rule engine first: banned words, exclamation marks, off-palette colours, contrast, logos on photos and the rules you’ve written. With an AI key set, a model review adds findings on top and reads images. Without one, you still get a real score from your real rules, and questions are answered from the rules where they can be. Nothing breaks because a model is down.
The full reference is in the docs, with every endpoint, error and limit. Start with the quickstart: one key, one config file, five minutes.
