# Heatshot: setup for agents

Heatshot predicts where people will look at a screen, an ad or a page, and returns
findings you can act on. You reach it as an MCP server or as a JSON API. Both need
the user's Heatshot key: ask them for it; never guess one.

## Connect over MCP

- Server: `https://api.heatshot.ai/mcp/` (streamable HTTP)
- Header: `Authorization: Bearer <key>`

Claude Code:

```bash
claude mcp add --transport http heatshot https://api.heatshot.ai/mcp/ --header "Authorization: Bearer <key>"
```

Other MCP clients:

```json
{"mcpServers": {"heatshot": {"type": "http", "url": "https://api.heatshot.ai/mcp/",
                             "headers": {"Authorization": "Bearer <key>"}}}}
```

## Or call the API

```bash
curl -s https://api.heatshot.ai/v1/analyze \
  -H "Authorization: Bearer $HEATSHOT_KEY" -H "content-type: application/json" \
  -d '{"image_url": "https://example.com/screen.png", "content_type": "web"}'
```

## How to use it

1. Analyze a screenshot (image) or a public page (address). Say which element should
   win attention with `target_text`, for example the main button's label.
2. Read `findings`: each has the measured value and the value that would clear it.
3. Change the design, analyze again, and repeat until the findings clear.
4. To choose between versions, compare a control with its variants.

Content types: web, mobile, ad, poster, packaging, shelf, ecommerce, thumbnail,
natural, serp, graphic-design, infographic.