AI Agents

Screenshot API for AI Agents

GetScreenshot gives AI agents eyes: a hosted screenshot API with an MCP server that returns images sized for vision models, from $5/month. Your agent sends a URL or a block of HTML and gets back an image it can look at, along with the estimated token cost.

claude mcp add --env GETSCREENSHOT_API_KEY=YOUR_API_KEY \
  --transport stdio getscreenshot -- npx -y getscreenshot-mcp

That is the install for Claude Code. Other clients are below.

Works with any API key on any plan. The first month of the $5 plan is free with code FREE5.

Why Agents Use a Hosted Screenshot API

Three reasons, with the numbers behind them

Works where there is no browser

Cloud coding agents, sandboxes and CI runners can't launch Chromium. GetScreenshot is an HTTPS call. It renders any public URL and hands back an image.

It also renders raw HTML. render_html takes up to 2 MB of markup the agent just wrote and returns a screenshot of it.

Output sized for models

The MCP server downsizes captures and slices full pages so each image fits a vision model's input budget. A 1280x800 viewport is one image of about 1,334 tokens. A full page comes back as slices of up to 1280x952, about 1,564 tokens each, four by default.

Every result reports its estimated token cost. The original full-resolution image stays available by URL.

Cheap and simple

One GET endpoint. Flat monthly plans from $5 for 2,500 screenshots, which is $0.002 per screenshot.

The MCP server, the CLI and the agent skill are included on every plan. There is no separate agent tier.

Three Ways to Connect

All three use the same API key and the same quota

1. MCP server

Five tools your agent can call directly. Images arrive inline in the tool result, already sized for the model.

npx -y getscreenshot-mcp

Runs locally over stdio with GETSCREENSHOT_API_KEY set, or connect to the hosted endpoint at https://mcp.getscreenshotapi.com/mcp.

MCP server reference

2. CLI and agent skill

One command from any shell. The getscreenshot skill teaches the agent when to run it, with no tool definitions in context.

npx -y getscreenshot-mcp capture https://example.com -o shot.png

Install the skill with npx skills add rasterwise/getscreenshot-mcp.

CLI and skill guide

3. Plain REST

One GET request. The response is JSON with screenshotImage, a link to the stored image.

curl "https://api.rasterwise.com/v1/get-screenshot?apikey=$GETSCREENSHOT_API_KEY&url=https%3A%2F%2Fexample.com&fullpage=true"

Machine-readable descriptions for agents that write their own client:

What Agents Use It For

Five jobs where looking at the page beats reading its HTML

Verifying UI changes after editing code

The agent edits a component, captures the preview URL, and checks the layout before it reports the task as done.

How the workflow looks

Reading a page visually

Charts, canvas elements and layout-dependent content don't survive HTML scraping. A screenshot shows the agent what a visitor sees.

Before and after comparison

compare_screenshots diffs two pages or two earlier captures. It returns the changed-pixel percentage, the bounding box of the change and a diff image.

Capturing evidence for reports

Every capture has a hosted URL that stays up for 30 days. The agent can paste it into a pull request, an issue or a report. capture_pdf saves a page as an A4 PDF.

Rendering HTML it just generated

render_html turns an HTML string into a screenshot. Useful for a component, an email template or a chart the agent wrote a moment ago.

Use case: visual verification for coding agents

The full workflow, with tool calls and a rules-file snippet you can paste into AGENTS.md or CLAUDE.md.

Read the use case

The Five MCP Tools

All read-only. Inputs and outputs are in the MCP server reference.

Tool What it does Quota used
capture_screenshot Captures a public web page as an image the model can see. Viewport, full page, a single element, or a device preset. 1
render_html Renders an HTML string of up to 2 MB and screenshots it. For a component or page the agent just wrote. 1
compare_screenshots Visual diff of two pages or two existing screenshots. Returns the changed-pixel percentage, the bounding box of the changed area and a diff image. 1 per page captured
capture_pdf Saves a page as an A4 PDF and returns a link that works for 30 days. 1
get_usage Screenshots used, monthly limit and reset date for the configured key. 1

When to Use Something Else

A hosted screenshot API has limits. These are the ones that matter for agents:

  • It cannot reach localhost or private networks. The capture runs on our servers. Use render_html, a public preview URL, or a tunnel.
  • It takes single-shot captures. It does not run multi-step, logged-in browsing sessions. That is what Playwright MCP and other browser automation tools are for.
  • Some sites refuse headless browsers. Pages behind aggressive bot protection may not load.
GetScreenshot vs Playwright MCP

Pricing

Flat monthly plans. The MCP server, CLI and agent skill work on all three.

Lincoln

$5/month

2,500 screenshots

$0.002 per screenshot

Hamilton

$10/month

6,000 screenshots

About $0.0017 per screenshot. Adds the Zapier, n8n and Make integrations.

Jackson

$20/month

15,000 screenshots

About $0.0013 per screenshot. Adds setup help.

There is no free tier. Code FREE5 makes the first month of Lincoln free, or takes $5 off the other plans. Your API key is emailed after checkout.

Questions

Does it work on the $5 plan?

Yes. The MCP server, the CLI and the agent skill work with any API key on any plan, including Lincoln at $5/month for 2,500 screenshots. They call the same REST API. Only the Zapier, Make and n8n integrations need Hamilton or higher.

Is there a free tier?

No. GetScreenshot has no free tier and no free trial. The code FREE5 makes the first month of the $5 Lincoln plan free, or takes $5 off the other plans.

Can it screenshot localhost?

No. The capture runs on GetScreenshot's servers, so it cannot reach localhost or a private network. Use render_html to render HTML the agent has in hand, or point the capture at a public preview URL or a tunnel to your dev server.

How many tokens does a screenshot cost?

With the MCP server, a default 1280x800 viewport capture is one image of about 1,334 tokens. A 390x844 phone viewport is about 434 tokens. A full page is cut into slices of up to 1280x952, about 1,564 tokens each, and four slices are returned by default. The estimate is ceil(width / 28) * ceil(height / 28) per image and is approximate. Every result reports it, and return: "url_only" skips the images entirely.

How long are images kept?

30 days. Each capture is stored on Google Cloud Storage at a public URL. Anyone who has the URL can open it, so do not capture pages that show data you would not put behind a public link.

Does a failed capture count?

Yes. Every authenticated request uses one unit of the monthly quota, including captures that fail and calls to get_usage. A selector that matches nothing or a page that never loads still counts, so have the agent fix the input before retrying.

Can it log in to a site or click through a flow?

No. Each call is a single capture of one URL. The click input clicks one selector before the capture, and that is the extent of the interaction. For multi-step or logged-in browsing, use a browser automation tool such as Playwright MCP.

Connect Your Agent

Get an API key, add the MCP server, and ask your agent to take a screenshot. Plans start at $5/month for 2,500 screenshots.