Let an AI agent see a web page
To let an AI agent see a web page, connect it to the CurlShot MCP server. The agent gets a take_screenshot tool: it passes the address of a page and receives a screenshot of it, which a model that reads images can look at. Agents without MCP use the same thing as one HTTP request, described in an OpenAPI file and in llms.txt.
curl "https://curlshot.com/api/v1/screenshot?access_key=YOUR_ACCESS_KEY&url=https://example.com&response_type=json&image_width=1200" \
--output screenshot.jsonHow to let an AI agent see a web page in three steps
- 1
Create a free account and copy your access key.
- 2
Add the MCP server to your agent, or give it the screenshot request.
- 3
Ask the agent to look at a page. It takes the screenshot itself.
Three ways in, for any kind of agent
Pick the one your agent already speaks. All three use the same access key and the same options.
Show the table (3 rows)
| Way in | Address | Best for |
|---|---|---|
| MCP server | /api/mcp, with the tools take_screenshot and list_devices | Claude, Cursor and other MCP clients |
| OpenAPI description | /openapi.json | Agents and code generators that build a client from a spec |
| Plain HTTP and llms.txt | /llms.txt, and the screenshot request | Any agent that can fetch a URL |
- Create a separate access key for the agent, so you can revoke it without touching your other keys.
- Leave "Require signature" off for that key: calls through MCP are not signed.
{
"mcpServers": {
"screenshots": {
"type": "http",
"url": "https://curlshot.com/api/mcp",
"headers": { "Authorization": "Bearer YOUR_ACCESS_KEY" }
}
}
}What people use it for
- A coding agent that checks how the page it just changed looks
- Research agents that read pages which need a real browser to load
- Visual checks: is the banner there, did the layout break
- Reports with screenshots, written by an assistant
For developers: one request
The same thing from your own code. It is a plain HTTP request, so it works from any language. Create a free account to get an access key.
Questions people ask
How can an AI agent take a screenshot of a website?
Through the CurlShot MCP server, which gives the agent a take_screenshot tool, or with one HTTP request that returns the image.
What does the agent get back?
The image itself, or with response_type=json a short JSON answer with a link to the file, which is easier for a tool call to pass on.
Can the agent read the page as text too?
The service returns images and PDFs. A PDF of a page has text that can be selected and extracted, and a model that reads images can read a screenshot directly.
How do I keep the cost of an agent under control?
Each plan is a fixed number of screenshots a month, and an access key can be revoked at any time. Repeat requests for the same page can come from the cache at no cost.
Related tools
- Screenshot APIA screenshot API for developers. One HTTP request turns a URL, HTML or Markdown into a PNG, JPEG, WebP or PDF. Full page, devices, blocking, cache, MCP.
- Full-page screenshotTake a screenshot of a whole web page, from the top to the very bottom, as one image. Paste the address, nothing to install. Also available as an API.
- HTML to imageRender your own HTML, CSS or Markdown into an image or a PDF with one API request. For social cards, certificates, receipts and reports.