Skip to main content
Bloom MCP Bloom is the brand layer. Add a brand once and its identity lives in Bloom as a working system that agents can call. See How Bloom works for the relationship between brands, Brand Skills, the API, and MCP. The Bloom MCP server connects Bloom to Claude and other AI agents through the Model Context Protocol (MCP). Once connected, you create images just by chatting — describe what you want and your agent makes it with Bloom, on your brand.

​
What you can do

  • Add a brand — use a website, Instagram account, files, or a combination of them, and everything you make after comes back on-brand.
  • Generate and edit images — make on-brand images from a prompt, then edit them, resize to any format, remove backgrounds, or vectorize. See Generate images and Edit and adapt images.
  • Bring your own images — upload photos or screenshots, or pull references from the Brand Library, to steer a generation.
  • Use reference ads — find an ad in Bloom’s curated library and recreate its composition for your brand. See Generate images.

​
Install the Bloom Skill (optional)

The Bloom Skill gives compatible agents persistent guidance about what Bloom is, when to use it, and how to choose between the API and MCP. It contains no brand data and does not connect or authenticate Bloom.
Install it in the current project so the guidance can be reviewed and shared with the repository:
Add --global when you deliberately want the guidance across unrelated projects.
You can skip this step and connect MCP directly. The public Skill source is inspectable before installation.

​
Connect your agent

Copy the Bloom MCP server URL:
The simplest way to connect is to sign in with Bloom — a browser sign-in (OAuth) on first connect, with nothing to set up in advance. Prefer a static key, or running somewhere without a browser? An API key works too.

Show setup steps for your agent →

Step-by-step setup for Claude, Codex, Cursor, ChatGPT desktop, and more — OAuth and API key.

​
How it works

Once connected, your agent discovers the Bloom tools and schemas currently available to the connected account. You drive them in plain language; there is no need to memorize tool names. Capabilities can vary by account during a rollout, so the live tool descriptions are the source of truth. Everything lives in a workspace (your personal one by default; ask your agent to list workspaces if you want to use another). A session follows one loop: Here’s your first run, end to end.

​
Onboard a brand

Give Bloom a website, Instagram account, or supporting files. The Brand Agent can combine those sources into one brand and publish an active Brand Skill containing its identity and guidance:
Your agent calls bloom_onboard_brand, then checks the returned brand with bloom_get_brand until it is ready. Already have brands in Bloom? Ask the agent to list them:
See Create a brand for supported sources, direct tool input, explicit-logo handling, failures, and compatibility inputs.

​
Generate your first image

Name the brand and describe what you want:
Your first brand comes with a few free credits, so you can generate as soon as the brand is ready. Bloom queues the image while your agent waits for the result and shows it inline — no copy-pasting URLs. Here’s what you can do next with the image you just generated:
  • Edit — change part of an image while keeping everything else.
  • Resize — adapt an image to a new aspect ratio for another channel.
  • Remove background — get a clean, transparent PNG.
  • Vectorize — turn a logo or icon into a scalable SVG.
See Edit and adapt images for all four.

​
Credits & plans

MCP access is available on all Bloom plans. Generation uses the same credit system as the API — see Credits and limits for plan tiers, per-image costs, and top-ups.