---
title: Example prompts
description: Plain-English prompts you can paste into Claude Code, Claude.ai, Cursor, or Codex once the Brand Stack MCP connector is authorized.
---

These are the same prompts surfaced on the [marketing page](https://brandstack.dev/mcp) and the
in-app **Settings → Developer** page, in copyable form. The agent picks the
right tool from `tools/list` and fills in the arguments; you don't need to
name the tool yourself.

Some prompts assume specific brand, persona, or product names from your
workspace ("Acme", "Sarah", "Operator"). Swap them for ids or names from
your own data.

## Read your workspace

```text
List all my brands and tell me which one has the most personality keywords.
```

```text
Pull my Acme brand as a Tailwind config snippet so I can paste it into my project.
```

```text
List the personas for my Operator product, then summarize the top three pain points across all of them.
```

```text
Find every mockup-generated asset from the past week and show me the URLs.
```

## Generate

`create_mockup` and `create_scene` consume credits from the connected
workspace (a new workspace's first 10 mockups are free). Generation
typically takes 20 to 60 seconds.

```text
Take a screenshot of the page we just built and turn it into a MacBook Pro mockup.
```

```text
Capture mysite.com at a mobile viewport with Brand Stack and turn it into an iPhone mockup.
```

```text
Create a 16:9 phone mockup featuring my persona Sarah, with natural lighting and an over-the-shoulder camera angle.
```

```text
Generate a 9:16 scene of persona Marcus working in a coffee shop with golden-hour lighting, for an Instagram story.
```

## Multi-step workflows

```text
Pick the most senior persona from my Acme product and create three scene variations of them in different environments.
```

```text
Read my latest Q4 strategy document and check if it mentions any pain points my personas care about.
```

## Server prompts (in your client's prompt picker)

The connector ships three **MCP prompts**, templates the server provides
and your client lists by name. In Claude they appear in the
`+` / slash-command menu once the connector is authorized:

| Prompt | What it does |
|--------|--------------|
| **Turn a screenshot into a mockup** | The headline flow. Finds the freshest screenshot in your session (a Playwright capture, a pasted image, a URL to capture), infers the device from its shape, and generates with at most one question. |
| **Create a mockup (guided)** | The deliberate version: walks through screenshot, saved setup, persona, device, aspect ratio, camera, and lighting, offering the studio's real options, then generates once you confirm. |
| **Create a scene (guided)** | Same, for a lifestyle scene: persona, environment, aspect ratio, camera, lighting. |

The guided prompts take an optional `subject` argument, what the image
is for ("a hero image for our pricing page"). The screenshot prompt
takes an optional `screenshot` argument, a path or URL, when you want to
point at a specific one.

You don't have to invoke any of them. The same rules ride in the tool
descriptions and the connector's session instructions, so an agent that
reaches for `create_mockup` on its own still infers what it can and asks
at most one question before spending your credits. The prompts just make
the flow deliberate and discoverable.

## Building your own

Once you've tried the examples, try open-ended prompts that combine the
read tools as input to a write tool. The agent loop is short:

1. Use the read tools (`list_*`, `get_*`) to gather context.
2. Get a screenshot in with `capture_screenshot` (a public URL), `create_upload_link` (a local file or a pasted image, no key needed), or the CLI upload.
3. Use a generation tool (`create_mockup`, `create_scene`) to act on it.
4. Use `list_assets` afterwards to confirm or fetch siblings.

See the [tools reference](/mcp/tools) for the full set of inputs each
tool accepts.