# GetScreenshot: full reference for LLMs and agents > GetScreenshot gives AI agents eyes: a hosted screenshot API with an MCP server that returns images sized for vision models, from $5/month. This file is a single, self-contained reference: the REST API, every parameter, responses and errors with fixes, worked examples, the MCP server, the CLI, the agent skill, limits and pricing. Product: GetScreenshot by Rasterwise, LLC. Site: https://www.getscreenshotapi.com Docs: https://docs.rasterwise.com Support: support@rasterwise.com Short version of this file: https://www.getscreenshotapi.com/llms.txt OpenAPI 3.1 description: https://www.getscreenshotapi.com/openapi.yaml ## 1. What it is GetScreenshot renders a public web page in a hosted headless Chromium browser and stores the result as a PNG, JPEG or PDF. You call one HTTPS GET endpoint and get back JSON with the URL of the stored file. ```bash curl "https://api.rasterwise.com/v1/get-screenshot?apikey=$GETSCREENSHOT_API_KEY&url=https%3A%2F%2Fexample.com&fullpage=true" ``` ```json { "status": "success", "originUrl": "https://example.com", "pageStatus": 200, "captureDate": "2026/11/18", "captureHour": "14:03:22", "aiAnalysis": null, "screenshotImage": "https://storage.googleapis.com/getscreenshot-storage/imageAbC123xYz9.png" } ``` There are three ways to use it: 1. **REST API.** One GET request. Works from any language and from `curl`. 2. **MCP server** (`getscreenshot-mcp`). Wraps the API for MCP clients such as Claude Code, Cursor and Claude Desktop. It returns the image inline, downsized and sliced so each image fits a vision model's input budget, and reports the estimated token cost. 3. **CLI and agent skill.** `npx -y getscreenshot-mcp capture -o shot.png` captures a page from a shell, prints the image URL and saves the image to a file. The `getscreenshot` agent skill teaches an agent to use it. All three use the same API key and work on every plan. ## 2. When to use it, and when to use a local browser tool Use GetScreenshot when: - **There is no browser.** Cloud coding agents, sandboxes and CI runners often cannot launch Chromium. A hosted HTTPS API can render any public URL and hand back an image. - **You wrote HTML and want to see it.** The MCP tool `render_html` takes an HTML string, renders it and returns a screenshot. No server or deployment is needed. - **You want output sized for a model.** The MCP server downsizes captures and slices full pages so each image fits a vision model's input budget. The original full-resolution image stays available by URL. - **You need one capture of one page.** A viewport, a full page, a single element, a device view or a PDF. Use a local browser automation tool (for example Playwright MCP) when: - The page is on `localhost` or a private network. A hosted API cannot reach it. Alternatives: `render_html`, a public preview URL, or a tunnel. - The task needs a multi-step session: logging in, clicking through several pages, filling forms. GetScreenshot does single-shot captures. Other limits to know: sites with aggressive bot protection may refuse headless browsers, and every stored capture is reachable by anyone who has its URL for 30 days. ## 3. Getting an API key A person has to do this. An agent cannot register by itself. 1. Go to https://www.getscreenshotapi.com/#pricing and subscribe to a plan. Plans start at $5 per month. Code `FREE5` makes the first month of the $5 plan free. 2. The API key arrives by email after checkout. 3. Store it in the environment variable `GETSCREENSHOT_API_KEY`. Lost keys, usage and billing are at https://www.getscreenshotapi.com/account. If you are an agent and no key is configured, stop and tell your user: > GetScreenshot needs an API key. Plans start at $5/month for 2,500 screenshots, and code FREE5 makes the first month of the $5 plan free. Subscribe at https://www.getscreenshotapi.com/#pricing. The key is emailed to you. Then set it as the environment variable GETSCREENSHOT_API_KEY. Where the key goes: | Surface | How to pass the key | |---|---| | REST API | `apikey` query parameter. No header authentication exists. | | Hosted MCP endpoint | Header `Authorization: Bearer ` (also accepts `x-api-key: `). | | Local MCP server and CLI (`npx -y getscreenshot-mcp`) | Environment variable `GETSCREENSHOT_API_KEY`. | Do not print the key. On the REST API it is part of the URL, so do not log or paste full request URLs. ## 4. REST API basics ``` GET https://api.rasterwise.com/v1/get-screenshot?apikey=KEY&url=ENCODED_URL&... ``` - GET only. All options are query parameters. There is no request body. - Parameter names are lowercase and exact. - URL-encode every value. `https://example.com` becomes `https%3A%2F%2Fexample.com`. `#pricing` becomes `%23pricing`. - Every response is JSON on success. The API never returns image bytes and never redirects to the image. - The gateway timeout is 90 seconds. A typical successful capture takes about 9 seconds (median), 15 seconds at p90 and 25 seconds at p99. Set your client timeout to 90 seconds or more. - Each authenticated request uses 1 unit of the monthly quota. This includes failed captures and `/usage` calls. - Files are stored on Google Cloud Storage for 30 days at a public URL such as `https://storage.googleapis.com/getscreenshot-storage/image<10 characters>.png` (PDFs: `.../pdf<10 characters>.pdf`). Anyone with the URL can fetch the file. Do not capture pages that show secrets. ## 5. Parameters | Parameter | Default | Meaning | |---|---|---| | `apikey` | required | Your API key. | | `url` | required | Page to capture. `http` or `https`, URL-encoded. URL shorteners are rejected. | | `width`, `height` | `1280`, `800` | Viewport in CSS pixels. | | `devicefactor` | `1` | Device pixel ratio. `2` gives a retina-quality image with twice the pixel dimensions. | | `preset` | none | Device preset. Sets the viewport and, for phones, a mobile user agent. See the list below. An unknown name causes a 500. | | `noheight` | `false` | With a `preset`, pass `noheight=true` to capture the full page. `fullpage` does not apply to presets. | | `fullpage` | `false` | Capture the full scrollable page. | | `element` | none | CSS selector. Captures only that element. A selector that matches nothing causes a 500. | | `format` | `png` | `png` or `jpeg`. | | `pdf` | `false` | Return an A4 PDF. The response has `pdfFile` instead of `screenshotImage`. | | `hidecookie` | `true` | Hides common cookie banners. Pass `false` to keep them. | | `hidemsg` | `false` | Hides common chat widgets (Intercom, Drift and similar). | | `hideelement` | none | CSS selector or selectors to hide (`display:none`) before capture. | | `customcss` | none | CSS injected before capture. | | `customjs` | none | JavaScript injected before capture. | | `highlight` | none | Keyword or keywords to highlight on the page. | | `click` | none | CSS selector to click before capture. | | `cookie` | none | `name,value` cookie set for the target host. | | `waituntil` | `networkidle2` | `load`, `domcontentloaded`, `networkidle0` or `networkidle2`. `domcontentloaded` is the fastest and the most reliable on heavy sites. | | `timewait` | `2000` | Extra wait in milliseconds before capture. | | `scrollmotion` | `true` | Scrolls the page first to trigger lazy-loaded content. | | `forcetr` | `false` | Transparent background where the page has none. | | `lang` | `en-US` | Accept-Language and browser language. | | `strategy` | `0` | Full-page capture strategy, 0 to 3. Try another value if a full-page capture renders badly. | | `email`, `emailsubject`, `emailmessage` | none | Emails the capture to an address, with an optional subject and message. | | `webhook` | none | URL that receives a POST with the result after capture. The HTTP call still waits for the capture. | | `urlasfilename` | `false` | Names the stored file after the URL. | Device presets (`preset`): `iphone5`, `iphone678`, `iphone678_plus`, `iphonex`, `iphone12`, `pixel2`, `pixel2_xl`, `ipad`, `ipadpro`, `hvga`, `wvga`, `dvga`, `wxga_v`, `xga`, `wxga_s`, `wxga_l`, `sxga`, `wsxga_plus`. Image formats are `png` and `jpeg` only. The API reference also documents `bplogin`, a form-login automation that needs credentials in the request. Agents should not use it: credentials should not pass through a language model. ## 6. Response Success is HTTP 200 with a JSON body: | Field | Type | Meaning | |---|---|---| | `status` | string | `"success"`. | | `originUrl` | string | The page that was captured. | | `pageStatus` | number | HTTP status the target page returned. Check it: a 404 here means you captured an error page. | | `captureDate` | string | UTC date, `YYYY/MM/DD`. | | `captureHour` | string | UTC time, `HH:MM:SS`. | | `aiAnalysis` | null | Legacy field. Ignore it. | | `screenshotImage` | string | Public URL of the image. Present for image captures. | | `pdfFile` | string | Public URL of the PDF. Present instead of `screenshotImage` when `pdf=true`. | | `emailSentStatus` | any | Present only when `email` was passed. | | `webhookPostStatus` | any | Present only when `webhook` was passed. | To get the file, send a plain GET to the URL in `screenshotImage` or `pdfFile`. No key is needed for that request. ## 7. Errors and how to fix them | HTTP | Body | Meaning | Fix | |---|---|---|---| | 401 | `{"message":"Unauthorized"}` | `apikey` query parameter missing. | Add `apikey=` to the query string. A header will not work. | | 500 | `{"message": "error"}` with response header `x-amzn-errortype: AuthorizerConfigurationException` | API key not recognised. This is the usual answer for an unknown key, a 500 for legacy reasons. | Check the key for typos. Do not retry with the same key. Recover the key at https://www.getscreenshotapi.com/account. | | 403 | `{"message":"Forbidden"}` | Key exists but is disabled (subscription cancelled or payment failed). An unrecognised key can also get this answer. | Check the key for typos. The account owner checks billing at https://www.getscreenshotapi.com/account. Do not retry. | | 429 | `{"message":"Limit Exceeded"}` | Monthly quota used up. | Stop. Wait for the reset date (see `/usage`) or upgrade the plan. | | 429 | `{"message":"Too Many Requests"}` | Rate limited. | Retry with exponential backoff. | | 400 | `{"status":"denied","message":"Bad posture."}` | `url` missing or not a valid http(s) URL. | Pass a full URL with scheme, URL-encoded. | | 403 | `{"status":"denied","message":"Access denied - URL not allowed (shortener or blocked domain)"}` | Blocked URL. | Use the final destination URL, not a shortener. | | 500 | `{"status":"failed","message":"Could not load the target page within the time limit. ..."}` | Page did not load: slow, offline, or blocking automated browsers. | Retry once with `waituntil=domcontentloaded` and a lower `timewait`. | | 500 | plain text `Internal Server Error` (not JSON) | Capture crashed. Most often the `element` selector was not found, the `preset` name is unknown, or the `click` selector is bad. | Fix or remove that parameter. Take a plain viewport capture first if you are unsure of the selector. | | 504 | `{"message": "error"}` | Took longer than 90 seconds. | Retry with `waituntil=domcontentloaded`, without `fullpage`, or with a lower `timewait`. | Rules for clients: - Check the HTTP status before parsing. A 500 body can be plain text. - Tell the two 403s, the two 429s and the three 500s apart by the body. - Failed captures still use quota. Change something before each retry. ## 8. Worked examples (curl) All examples read the key from `GETSCREENSHOT_API_KEY`. Viewport at a custom size: ```bash curl "https://api.rasterwise.com/v1/get-screenshot?apikey=$GETSCREENSHOT_API_KEY&url=https%3A%2F%2Fexample.com&width=1440&height=900" ``` Full page: ```bash curl "https://api.rasterwise.com/v1/get-screenshot?apikey=$GETSCREENSHOT_API_KEY&url=https%3A%2F%2Fexample.com&fullpage=true" ``` One element (`#pricing`, encoded as `%23pricing`): ```bash curl "https://api.rasterwise.com/v1/get-screenshot?apikey=$GETSCREENSHOT_API_KEY&url=https%3A%2F%2Fexample.com&element=%23pricing" ``` Mobile device preset, viewport only and then full page: ```bash curl "https://api.rasterwise.com/v1/get-screenshot?apikey=$GETSCREENSHOT_API_KEY&url=https%3A%2F%2Fexample.com&preset=iphone12" curl "https://api.rasterwise.com/v1/get-screenshot?apikey=$GETSCREENSHOT_API_KEY&url=https%3A%2F%2Fexample.com&preset=iphone12&noheight=true" ``` PDF (read `pdfFile` from the response): ```bash curl "https://api.rasterwise.com/v1/get-screenshot?apikey=$GETSCREENSHOT_API_KEY&url=https%3A%2F%2Fexample.com&pdf=true" ``` Hide elements. Cookie banners are hidden by default; this also hides chat widgets and two custom selectors (`.promo-banner, #newsletter-modal`): ```bash curl "https://api.rasterwise.com/v1/get-screenshot?apikey=$GETSCREENSHOT_API_KEY&url=https%3A%2F%2Fexample.com&hidemsg=true&hideelement=.promo-banner%2C%20%23newsletter-modal" ``` Wait strategies. For a slow or script-heavy page, stop waiting on the network and give the page 4 seconds to render: ```bash curl "https://api.rasterwise.com/v1/get-screenshot?apikey=$GETSCREENSHOT_API_KEY&url=https%3A%2F%2Fexample.com&waituntil=domcontentloaded&timewait=4000" ``` For a page that must finish every network request first: ```bash curl "https://api.rasterwise.com/v1/get-screenshot?apikey=$GETSCREENSHOT_API_KEY&url=https%3A%2F%2Fexample.com&waituntil=networkidle0" ``` Let curl do the encoding, keep the HTTP status, then download the image: ```bash status=$(curl -sS --max-time 100 -o result.json -w '%{http_code}' -G \ "https://api.rasterwise.com/v1/get-screenshot" \ --data-urlencode "apikey=$GETSCREENSHOT_API_KEY" \ --data-urlencode "url=https://example.com/pricing" \ --data-urlencode "element=.pricing-table") echo "HTTP $status" # On 200 only (needs jq): curl -sS -o shot.png "$(jq -r .screenshotImage result.json)" ``` ## 9. Python example ```python import os import requests API = "https://api.rasterwise.com/v1/get-screenshot" def screenshot(url: str, **options) -> str: """Capture `url` and return the public URL of the stored image or PDF.""" params = {"apikey": os.environ["GETSCREENSHOT_API_KEY"], "url": url, **options} resp = requests.get(API, params=params, timeout=100) # requests URL-encodes params if resp.status_code != 200: # Error bodies are usually JSON, but a crashed capture returns plain text. raise RuntimeError(f"GetScreenshot HTTP {resp.status_code}: {resp.text[:300]}") data = resp.json() return data.get("screenshotImage") or data["pdfFile"] image_url = screenshot("https://example.com", fullpage="true") with open("shot.png", "wb") as f: f.write(requests.get(image_url, timeout=60).content) ``` ## 10. Node.js example (fetch) Runs on Node.js 18 or later with no dependencies. ```js import { writeFile } from "node:fs/promises"; const API = "https://api.rasterwise.com/v1/get-screenshot"; async function screenshot(url, options = {}) { const params = new URLSearchParams({ apikey: process.env.GETSCREENSHOT_API_KEY, url, ...options, }); // URLSearchParams URL-encodes every value const res = await fetch(`${API}?${params}`, { signal: AbortSignal.timeout(100_000) }); const text = await res.text(); if (!res.ok) { // Error bodies are usually JSON, but a crashed capture returns plain text. throw new Error(`GetScreenshot HTTP ${res.status}: ${text.slice(0, 300)}`); } const data = JSON.parse(text); return data.screenshotImage ?? data.pdfFile; } const imageUrl = await screenshot("https://example.com", { fullpage: "true" }); const image = await fetch(imageUrl); await writeFile("shot.png", Buffer.from(await image.arrayBuffer())); ``` ## 11. Other endpoints ### GET /usage ```bash curl "https://api.rasterwise.com/v1/usage?apikey=$GETSCREENSHOT_API_KEY&email=me%40example.com" ``` ```json { "usedQuota": 123, "remainingQuota": { "limit": 6000, "offset": 0, "period": "MONTH" }, "quotaRestartDate": "2026-12-01" } ``` - `remainingQuota` is the plan quota object, not a remaining count. `limit` is the monthly allowance. Remaining requests = `remainingQuota.limit - usedQuota`. - `email` must be present with any non-empty value. Without it the endpoint returns 400 `{"message":"Email parameter is required"}`. - It can return 400 `{"message":"Invalid API key"}`. - Calling it uses 1 quota unit. ### GET /get-screenshot/validate-key ``` GET https://api.rasterwise.com/v1/get-screenshot/validate-key?apikey=KEY ``` Returns 200 `{"valid":true,"email":"..."}` or 403 `{"valid":false}`. This endpoint exists for the Zapier, Make and n8n integrations. It returns 403 for Lincoln-plan keys even though those keys are valid for the API, so do not use it as a general key check. To test a key, make a normal `/get-screenshot` or `/usage` call. ## 12. MCP server The MCP server is the npm package `getscreenshot-mcp`. Server name: `getscreenshot`. Title: "GetScreenshot". It calls the REST API with your key, so it works with any API key on any plan. ### Install Run locally over stdio: ```bash GETSCREENSHOT_API_KEY=YOUR_API_KEY npx -y getscreenshot-mcp ``` Or connect to the hosted endpoint over Streamable HTTP: ``` https://mcp.getscreenshotapi.com/mcp Authorization: Bearer YOUR_API_KEY ``` The hosted endpoint also accepts the header `x-api-key: YOUR_API_KEY`. Claude Code, hosted endpoint: ```bash claude mcp add --transport http getscreenshot https://mcp.getscreenshotapi.com/mcp --header "Authorization: Bearer YOUR_API_KEY" ``` Claude Code, local server: ```bash claude mcp add --env GETSCREENSHOT_API_KEY=YOUR_API_KEY --transport stdio getscreenshot -- npx -y getscreenshot-mcp ``` Generic JSON configuration (Cursor, Claude Desktop and other clients): ```json { "mcpServers": { "getscreenshot": { "command": "npx", "args": ["-y", "getscreenshot-mcp"], "env": { "GETSCREENSHOT_API_KEY": "YOUR_API_KEY" } } } } ``` Per-client setup guides: https://www.getscreenshotapi.com/agents/ When no API key is configured, every tool returns an error result that explains how to get a key (https://www.getscreenshotapi.com/#pricing, plans from $5/month, code FREE5) and how to set it. ### Tools All five tools are read-only (`readOnlyHint: true`, `destructiveHint: false`, `openWorldHint: true`). **`capture_screenshot`**: capture a public web page as an image the model can see. | Input | Type | Maps to | Notes | |---|---|---|---| | `url` | string, required | `url` | Public http or https URL. | | `full_page` | boolean, default false | `fullpage` | Full scrollable page. | | `element` | string | `element` | CSS selector. | | `device` | enum | `preset` | One of the device presets in section 5. | | `width`, `height` | integer | `width`, `height` | Viewport in CSS pixels. | | `scale` | 1 to 3 | `devicefactor` | Device pixel ratio. | | `format` | `png` or `jpeg`, default `png` | `format` | Format of the stored file. | | `wait_until` | enum | `waituntil` | `load`, `domcontentloaded`, `networkidle0`, `networkidle2`. | | `wait_ms` | 0 to 20000 | `timewait` | Extra wait before capture. | | `hide_selectors` | string | `hideelement` | CSS selector or selectors to hide. | | `hide_cookie_banners` | boolean, default true | `hidecookie` | | | `hide_chat_widgets` | boolean | `hidemsg` | | | `click` | string | `click` | CSS selector to click first. | | `custom_css`, `custom_js` | string | `customcss`, `customjs` | Injected before capture. | | `highlight` | string | `highlight` | Keyword to highlight. | | `cookie` | `{name, value}` | `cookie=name,value` | | | `lang` | string | `lang` | | | `transparent_background` | boolean | `forcetr` | | | `full_page_strategy` | 0 to 3 | `strategy` | | | `detail` | `standard` or `high`, default `standard` | | Image sizing for the model. See below. | | `max_slices` | 1 to 10, default 4 | | Maximum number of inline images for a tall page. | | `return` | `image_and_url` (default) or `url_only` | | `url_only` skips the inline images. | **`render_html`**: render an HTML string, for example a component or page the agent just wrote, and screenshot it. Inputs: `html` (required, up to 2 MB) plus `full_page`, `element`, `device`, `width`, `height`, `scale`, `wait_ms`, `detail`, `max_slices` and `return`. The server hosts the HTML at a temporary, unguessable URL, captures it through the normal API, deletes the document as soon as the capture returns, and returns the same output as `capture_screenshot`. The page runs in a sandbox without cookies or localStorage, so code that depends on them may render blank. Relative asset URLs will not resolve unless the HTML sets `` or inlines its assets. **`compare_screenshots`**: visual diff of two pages or two existing screenshots, for example before and after a change, or staging against production. Inputs: `before` and `after`, each either `{url}` (a page to capture) or `{screenshot_url}` (a previous GetScreenshot result URL); shared `full_page`, `width`, `height`, `device`, `hide_selectors`, `wait_until`, `wait_ms`; and `threshold` (0 to 1, default 0.1). Output: the percentage of changed pixels, the bounding box of the changed area, an inline diff image with the changes highlighted, and both source URLs. It costs one screenshot per `{url}` input. At most the top 2,400 px of each side are compared, at up to 1280 px wide; to compare a region further down a long page, capture it with `element` first and pass the two results as `{screenshot_url}`. **`capture_pdf`**: save a page as an A4 PDF. Inputs: `url`, `wait_until`, `wait_ms`, `hide_selectors`, `custom_css`. Returns `pdf_url`, a 30-day link. **`get_usage`**: screenshots used, monthly limit and reset date for the configured key. Costs 1 quota unit. ### Output contract `capture_screenshot` and `render_html` return: 1. A text summary. 2. One or more inline JPEG image blocks, sized for vision models. 3. `structuredContent`: ```json { "screenshot_url": "https://storage.googleapis.com/getscreenshot-storage/imageAbC123xYz9.png", "origin_url": "https://example.com", "page_status": 200, "captured_at": "2026-11-18T14:03:22Z", "format": "png", "original": { "width": 1280, "height": 1904 }, "images": [ { "index": 0, "y": 0, "width": 1280, "height": 952, "estimated_tokens": 1564 }, { "index": 1, "y": 952, "width": 1280, "height": 952, "estimated_tokens": 1564 } ], "total_slices": 2, "truncated": false, "estimated_image_tokens": 3128, "expires_in_days": 30, "warnings": [] } ``` The values above are illustrative. `screenshot_url` is the original full-resolution file. `images` describes each inline image: its position in the page (`y`), its size and its approximate token cost. `truncated` is true when the page needed more slices than `max_slices` allowed. `original` is `null` if the stored image could not be read. `warnings` lists notes about the result, for example that the page answered with an HTTP error status. ### Image sizing rules (`detail`) | `detail` | Max width | Max long edge per image | Max estimated tokens per image | |---|---|---|---| | `standard` (default) | 1280 px | 1568 px | about 1568 | | `high` | 1920 px | 2576 px | about 4784 | - A full page is cut into slices so every image stays inside these limits. At `standard`, a 1280 px wide page is cut into slices about 952 px tall. - The token estimate is `ceil(width / 28) * ceil(height / 28)`. It is approximate and follows Claude-style image accounting. - Inline images are JPEG at quality about 80. The stored original keeps the `format` you asked for. - To spend fewer tokens: capture an `element` instead of a page, keep `detail` at `standard`, lower `max_slices`, or use `return: "url_only"`. ## 13. CLI The same npm package is a command-line tool. It reads the key from `GETSCREENSHOT_API_KEY`. ```bash npx -y getscreenshot-mcp capture "https://example.com" -o shot.png ``` ```bash # One element npx -y getscreenshot-mcp capture "https://example.com" --element "header nav" -o nav.png # Mobile device npx -y getscreenshot-mcp capture "https://example.com" --device iphone12 -o mobile.png # Full page npx -y getscreenshot-mcp capture "https://example.com" --full-page -o full.png # Slow page npx -y getscreenshot-mcp capture "https://example.com" --wait-until domcontentloaded --wait-ms 4000 -o app.png ``` | CLI flag | API parameter | |---|---| | `` | `url` | | `--full-page` | `fullpage` | | `--element SEL` | `element` | | `--device NAME` | `preset` | | `--width N`, `--height N` | `width`, `height` | | `--format png\|jpeg` | `format` | | `--wait-until X` | `waituntil` | | `--wait-ms N` | `timewait` | | `--hide SEL` | `hideelement` | | `-o FILE` | Also download the full-resolution file to this path. Without it the command only prints the URL. | | `--json` | Print the result as JSON, including the public screenshot URL. | The package also has `pdf ` (A4 PDF) and `usage` (screenshots used and left; costs 1 quota unit). Exit codes: 0 success, 1 the capture or download failed, 2 bad arguments (for example an unknown `--device` name, which is rejected before any capture). After the command exits, open the saved file with your image viewing tool. Do not claim a page looks right without viewing the image. ## 14. Agent skill The agent skill is named `getscreenshot`. It teaches a shell-only agent to take screenshots with the CLI or `curl`, pick the smallest capture that answers the question, and verify UI changes before and after. ```bash npx skills add rasterwise/getscreenshot-mcp ``` Source repository: https://github.com/rasterwise/getscreenshot-mcp Guide: https://www.getscreenshotapi.com/agents/agent-skill.html If the MCP tool `capture_screenshot` is available in the session, prefer it over the skill: it returns images already sized for vision models. ## 15. Limits - **Public URLs only.** The capture runs on GetScreenshot's servers. `localhost`, private networks and VPN-only hosts are unreachable. Use `render_html`, a public preview URL or a tunnel. - **Single-shot captures.** One page load per call. No multi-step, logged-in browsing sessions. - **Bot protection.** Sites with aggressive bot protection may refuse headless browsers. - **Timeout.** 90 seconds per request. - **Formats.** PNG or JPEG images. PDFs are A4. - **Storage.** Files are kept for 30 days at a public URL. Anyone with the URL can fetch them. - **Quota.** Every authenticated request counts, including failed captures and usage checks. - **Rate limits.** 5 requests per second on Lincoln, 10 on Hamilton and 10 on Jackson. - **URL shorteners** and some blocked domains are rejected with a 403. ## 16. Pricing | Plan | Price per month | Screenshots per month | Price per screenshot | Rate limit | Adds | |---|---|---|---|---|---| | Lincoln | $5 | 2,500 | $0.002 | 5 requests/second | REST API, MCP server, CLI, agent skill | | Hamilton | $10 | 6,000 | about $0.0017 | 10 requests/second | Zapier, n8n and Make integrations | | Jackson | $20 | 15,000 | about $0.0013 | 10 requests/second | Setup help | - Every plan is paid. Code `FREE5` gives the first month of Lincoln free, or $5 off the other plans. - The REST API, MCP server, CLI and agent skill work on every plan. Zapier, Make and n8n need Hamilton or higher. - Subscribe at https://www.getscreenshotapi.com/#pricing. Manage the account at https://www.getscreenshotapi.com/account. - Markdown version of this section: https://www.getscreenshotapi.com/pricing.md ## 17. Support and links - Support: support@rasterwise.com - API reference: https://docs.rasterwise.com/docs/getscreenshot/api-reference-0/ - Getting started: https://docs.rasterwise.com/docs/getscreenshot/getting-started/ - MCP server docs: https://docs.rasterwise.com/docs/agents/mcp-server/ - Agent skill docs: https://docs.rasterwise.com/docs/agents/agent-skill/ - LLM quickstart: https://docs.rasterwise.com/docs/agents/llm-quickstart/ - AI agents hub: https://www.getscreenshotapi.com/agents/ - Authentication for agents: https://www.getscreenshotapi.com/auth.md - OpenAPI description: https://www.getscreenshotapi.com/openapi.yaml - API catalog: https://www.getscreenshotapi.com/.well-known/api-catalog - MCP server card: https://www.getscreenshotapi.com/.well-known/mcp/server-card.json