localscreenshot - Screenshot API & MCP Server

Take a screenshot of any public web page from 227 countries, with cookie banners and ads hidden. Use it from your code with the REST API, or from your AI assistant through the Model Context Protocol.

This repository is the public documentation of the hosted service: the MCP endpoint, the tools, the client setups and the registry manifest. The service itself runs at localscreenshot.com, the command line client lives in localscreenshot-cli.

Quick start, no API key

Any HTTP MCP client can connect to the free endpoint. Real screenshots, no sign-up:

{
  "mcpServers": {
    "localscreenshot": {
      "type": "http",
      "url": "https://api.localscreenshot.com/mcp/free"
    }
  }
}

It answers initialize, tools/list, prompts/list and resources/list without credentials, and gives three tools: a real screenshot of one of three example pages, the list of countries, and the prices. It cannot capture a page of your choice.

As a plugin

The repository follows the Open Plugins standard, so directory scans find it: plugin.json carries the identity, mcp.json declares the connection. Both point at the free endpoint above, which means an install works immediately and gives the three demo tools listed below.

To capture a page of your choice through the same plugin, replace the url and add the headers block in mcp.json with the keyed setup in the next section. Nothing else changes, the tools are the same ones, they just stop being limited to the example pages.

With your own key

A free account gets 50 credits every month, no card: localscreenshot.com/signup. Create the key in the dashboard, under Connect.

{
  "mcpServers": {
    "localscreenshot": {
      "type": "http",
      "url": "https://api.localscreenshot.com/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_API_KEY"
      }
    }
  }
}

Claude Code, in one line:

claude mcp add --transport http localscreenshot https://api.localscreenshot.com/mcp --header "Authorization: Bearer YOUR_API_KEY"

Setups for each tool, written from that tool's own documentation: localscreenshot.com/mcp and docs.localscreenshot.com/agent.

Tools

With a key:

Tool	What it does
`take_screenshot`	Takes the screenshot. Returns the image, the page title, its final address and status, and the credits left.
`list_countries`	Lists the countries a screenshot can be taken from.
`get_credits`	Says how many credits are left.

On the free endpoint:

Tool	What it does
`take_demo_screenshot`	A real screenshot of one of three example pages, in one of two devices.
`list_countries`	The same country list, no key needed.
`get_pricing`	The credit costs of the service.

take_screenshot accepts the same options as the API, except async and response: the server waits for the picture and decides how to hand it back. The image is scaled down to 1,568 pixels wide unless you ask for another width, which keeps it readable for a vision model without spending tokens for nothing. A PDF, or an image over 4 MB, comes back as a short-lived download link instead of an attachment. Errors arrive as a plain sentence with their code.

Options

take_screenshot exposes the 28 parameters of the API, minus async and response: the server waits for the picture and decides how to hand it back.

Parameter	Type	What it does
`url`	string	The public http or https page to capture.
`country`	string	Load the page from this country, an ISO 3166-1 alpha-2 code. Costs 2 credits instead of 1. The response reports the country actually observed. If it cannot be obtained, the request fails and is not charged.
`viewport`	string	Screen size as `WIDTHxHEIGHT`, for example `1440x900`. Shortcut for the two below.
`viewport_width`	integer	Width of the browser window, in pixels.
`viewport_height`	integer	Height of the browser window, in pixels. With `full_page`, the image is taller than this.
`full_page`	boolean	Capture the whole scrollable page instead of the visible part.
`device`	string	Capture as this device, meaning screen size, pixel density and mobile browser. One of `desktop`, `desktop_hd`, `laptop`, `ipad`, `ipad_pro`, `ipad_mini`, `iphone_15`, `iphone_15_pro_max`, `iphone_se`, `pixel_8`, `galaxy_s24`.
`orientation`	string	Hold a phone or a tablet upright (`portrait`) or on its side (`landscape`). Needs a phone or tablet device.
`device_scale_factor`	number	Pixel density. `2` gives a sharper, larger image.
`format`	string	`png`, `jpeg`, `webp`, or `pdf` which saves the whole page as a document.
`quality`	integer	Quality for `jpeg` and `webp`. Ignored for `png` and `pdf`.
`selector`	string	Capture only the first element matching this CSS selector.
`max_width`	integer	Scale the image down to this width. Useful to keep it small for a vision model.
`click_text`	string	Click one visible button, tab or other clickable element, pointed at by the text a visitor reads, for example "Show all". One click, no typing. Costs 1 more credit.
`click_selector`	string	The same click, pointed at by a CSS selector matching exactly one element. Use it when you know the code of the page. One click, no typing. Costs 1 more credit.
`wait_for`	string	Wait until an element matching this CSS selector is visible.
`delay_ms`	integer	Extra wait before the capture, in milliseconds.
`block_cookie_banners`	boolean	Removes most cookie banners. A banner the filters do not know stays in the picture.
`block_ads`	boolean	Blocks ads.
`block_chat_widgets`	boolean	Hides chat bubbles.
`close_popups`	boolean	Closes pop-ups and newsletter overlays.
`hide_sticky_elements`	boolean	Hides headers, bars and buttons that stay fixed on screen.
`hide_selectors`	array	Hides every element matching these CSS selectors.
`disable_animations`	boolean	Freezes animations and transitions.
`color_scheme`	string	Ask the page for its `light` or `dark` theme.
`language`	string	Browser language, for example `fr` or `fr-FR`. Without it, a screenshot from a country uses the main language of that country.
`timezone`	string	Browser time zone, for example `Europe/Paris`. Without it, a screenshot from a country uses the main time zone of that country.
`user_agent`	string	User-Agent the page receives. A preset (`googlebot`, `googlebot_mobile`) or your own string. Useful to compare what a site shows to a crawler and to a visitor. A site that checks the crawler's address is not fooled.

Full details: Take a screenshot.

What it costs

1 credit for a screenshot, 2 from a chosen country. A failed screenshot is not charged.
50 free credits every calendar month, no card. Unused credits do not carry over.
A free account can send 10 requests a minute and run 2 screenshots at a time.
Plans and credit packs: localscreenshot.com/pricing.

Countries

227 countries, from AD Andorra onward. The list is served by GET https://localscreenshot.com/v1/countries, documented on Countries and attestation. The response of a country-specific capture reports the country actually observed, and a capture that could not be obtained from that country fails and is not charged.

REST API

curl -H "Authorization: Bearer YOUR_API_KEY" \
  "https://localscreenshot.com/v1/screenshot?url=https://example.com"

The response is the image. Add response=json for a link to the image plus the page details and the attestation. POST with a JSON body works the same way. Background captures return an id, read it at /v1/screenshots/{id}.

OpenAPI: localscreenshot.com/v1/openapi.json
Documentation: docs.localscreenshot.com/overview
For AI tools: llms.txt

Command line

npx localscreenshot https://example.com --full-page
pipx run localscreenshot https://example.com --country BR

Both packages are built from localscreenshot-cli, with no dependency and no telemetry. The key comes from LOCALSCREENSHOT_API_KEY or from localscreenshot login.

What it does not do

It cannot sign in to a website, solve a CAPTCHA or get past a paywall. It captures public pages only.

License

MIT. See LICENSE.

sambel/localscreenshot