# ChatGPT design: making on-brand designs from ChatGPT

[Canonical HTML page](https://gradiently.design/guide/design-with-chatgpt)

ChatGPT can draw you a picture. Connected to a design tool, it can do something more useful: make editable, on-brand designs you can open, fix and post.

## The short version

- ChatGPT design work is most reliable when ChatGPT uses a real design tool through an MCP connection rather than generating a flat image.
- Whether you can add a custom MCP connector in ChatGPT depends on your plan and settings, which OpenAI has been changing, so check its current help pages.
- Gradiently's hosted MCP endpoint is https://gradiently.design/api/mcp, and it authenticates with a Bearer API key created in Gradiently's Settings.
- Through MCP, an assistant can search Marks, read your brand, create designs that open in the Gradiently Studio and edit their text.
- A scoped API key limits what the assistant can do, and it can be revoked at any time.

**ChatGPT design** gets genuinely useful when ChatGPT stops drawing pictures and starts operating a design tool. Ask it for an image and you get a flat file with invented type and no way to change one word. Connect it to a tool like Gradiently through MCP, and the same request produces a real design: your fonts, your colours, your Mark, text that stays text, at an exact platform size, saved where you can open it and fix anything before you post.

This guide explains the difference, what you need for the connection, and how to ask for designs once it works. If MCP is new to you, [what is MCP](https://gradiently.design/guide/what-is-mcp) explains it in plain words.

## Image generation versus a connected design tool

Both have their place. Image generation is excellent for illustrations, concepts and moodboards. For the posts, stories and posters a brand publishes every week, a design tool behind the chat wins on almost every practical point.

### An image generated in chat

- A flat picture: every change means generating again
- Type drawn as pixels, sometimes misspelled
- Sizes approximate; you crop afterwards
- Brand colours and fonts described, not applied
- Hard to keep a series consistent

### A design made through MCP

- An editable design with layers and pages
- Real text in your brand fonts, exactly as written
- Exact sizes such as 1080 × 1350 or 1080 × 1920
- Your brand kit and Mark applied by the tool
- Every design in the series wears the same Mark

## What you need to connect ChatGPT to Gradiently

The connection has two ends. On the Gradiently side, you need an API key. On the ChatGPT side, you need a way to add a custom MCP server, sometimes called a connector, that sends that key with every request.

> **Check your ChatGPT plan and settings** Support for custom MCP connectors in ChatGPT depends on your plan and workspace settings, and OpenAI has been changing it. Check OpenAI's current help pages for adding a custom connector, and confirm it can send an API key in the Authorization header. If your ChatGPT account can't, any MCP client that can send a header will work with the same key, including Claude; see [designing with Claude](https://gradiently.design/guide/design-with-claude).

1. **Create a key in Gradiently** Go to **Settings** and open **API and agents**. Choose **Create a key**, name it after where it will be used, such as *ChatGPT*, and tick only the permissions it needs: usually **Read designs**, **Create and edit designs** and **Search Marks**. Only workspace owners and admins can create keys.
2. **Copy the key once** It is shown only once. Keep it in a password manager, never in a shared document.
3. **Add the MCP server in ChatGPT** Where your ChatGPT settings allow a custom connector, add the endpoint below and set the key as a Bearer token in the Authorization header.
4. **Check the tools appear** Once connected, the assistant can list Gradiently's tools. Ask it what it can do in Gradiently to confirm.

```text
Server URL:     https://gradiently.design/api/mcp
Authorization:  Bearer YOUR_GRADIENTLY_KEY
```

The two values any MCP client needs. Every request must carry the key; there is no anonymous access.

The full walkthrough for every client, including what each permission allows, is in [connecting an AI assistant](https://gradiently.design/guide/connect-ai-assistant). The protocol itself is documented at [modelcontextprotocol.io](https://modelcontextprotocol.io).

## What ChatGPT can do once connected

| Tool | What it lets the assistant do |
| --- | --- |
| `search_marks`, `get_mark` | Find Marks in the Market and read their recipes |
| `get_brand_profile` | Read what your brand is, who it is for and how it sounds |
| `list_templates`, `create_design` | Start a design from a template at a chosen size, with your copy, returning a Studio link |
| `compose_design`, `edit_design` | Lay out a design itself, get review notes on margins and contrast, and make small changes |
| `update_design_text` | Change the words of a saved design without touching its layout |
| `design` | Hand the brief to Gradiently's own Designer, which uses Gradiently AI credits |
| `make_mark`, `save_mark` | Forge a new Mark recipe and save it as your private draft |
| `render_design` | Render a saved design as PNG or PDF, if you own the Mark it wears |

Some of Gradiently's MCP tools. Which ones work depends on the permissions you gave the key.

When the assistant composes a design itself, the work runs on your own AI plan and spends no Gradiently credits. Every design it makes is saved in your workspace and comes back with a link that opens it in the Studio.

## How to ask for on-brand designs

Ask the way you would brief a person: what it is for, what it must say, where it will be posted. The assistant fills in the rest from your brand profile and kit, so it pays to set those up first; see [setting up your brand kit](https://gradiently.design/guide/brand-kit-setup).

A portrait event poster for a jazz night with the day, time and venue on a living gradient

*A poster for Thursday's jazz night, 8pm at the Blue Room, free entry.*

The same jazz night poster laid out as a vertical story on the same Mark

*Now the same as an Instagram story.* Same Mark and type, laid out again for 9:16.

- **Name the format:** *an Instagram post*, *a LinkedIn banner*, *an A4 poster*. The tool uses exact sizes.
- **Give the real details:** dates, times, prices, addresses. Invented details are the most common thing you'll have to fix.
- **Ask for series, not singles:** *five posts for launch week on the same Mark* keeps them consistent.
- **Open before you post:** follow the Studio link, check every word, and export from there.

## Keeping the connection safe

An API key is a password for your workspace, so treat it like one. Give it the fewest permissions that do the job: leave out **Claim Marks** unless you want the assistant to claim Marks for you, and **Generate brands** unless you are using it to create new brands. Each key shows when it was last used. If a key is exposed, or you stop using ChatGPT, revoke it under **API and agents** and anything using it stops working straight away. A key never changes who owns a Mark or who holds its licence; exporting still follows the usual rule that the Mark must be yours.

## FAQ

### Can ChatGPT make designs in Gradiently?

Yes, if your ChatGPT account can add a custom MCP connector that sends an API key. It can then search Marks and create designs that open in the Gradiently Studio.

### What is Gradiently's MCP endpoint?

The hosted endpoint is https://gradiently.design/api/mcp. Every request needs a Bearer API key created under Settings, API and agents.

### Does designing through ChatGPT use Gradiently credits?

When the assistant composes designs itself, it runs on your own AI plan and spends no Gradiently credits. Handing a brief to Gradiently's own Designer through the design tool does use credits.

### Is it safe to give ChatGPT an API key?

Give the key only the permissions it needs, keep it out of shared documents, and revoke it in Settings if it is exposed. Revoking stops it immediately.

### Why not just ask ChatGPT to generate an image?

A generated image is flat, with drawn type and approximate sizes. A design made through a design tool is editable, uses your real fonts and colours, and is made at exact platform sizes.