Skip to main content
Artbucket keeps a brand’s files and its rules in one place, and serves both to whoever asks: a person in the app, a script with an API key, or an agent in Claude, ChatGPT, Cursor or Figma. Finding files is a solved problem. The hard part now is authority: which logo is current, which photo is still licensed for paid social in Germany, what the brand’s actual rules are. No agent can infer that from pixels. It has to be written down and served, and that is what Artbucket does.

Quickstart

Running on your machine in five minutes.

How it works

The API, the MCP server and the canon.

Install it

Docker Compose, Docker, Fly, Coolify or a plain VPS.

Connect an agent

One URL, a consent screen, no pasted keys.

​
What makes it different

  • Renditions are pure functions of a URL. /a/{id}/w_1200,f_webp is the asset at that size. No export step, no download button.
  • The web app is a client of the public API. It has no private endpoints: anything it does, a script or an agent can do.
  • Content-addressed storage. Upload the same bytes twice, get one asset.
  • Your metadata stays yours. Title, credit and tags are written back into the file on download, so leaving costs nothing.
  • The brand as data. Colors, logo rules, type scale and tone are records an agent can query, and the guidelines page is drawn from them.
  • A yes or no on a use. POST /api/v1/check says whether an asset may run here, now, in this context, and what to use instead.
Artbucket is open source under the AGPL-3.0. Run it, change it, host it yourself. The API is not stable until v1.0.