GetScreenshot vs Playwright MCP
Hosted screenshots vs a local browser. Playwright MCP drives a real browser on your machine. GetScreenshot renders pages on its own servers and returns images sized for vision models. They do different jobs, and many setups use both.
Use Playwright MCP when
The agent has to do things in a browser: click through a flow, fill in a form, stay logged in, or work against localhost. It is free and open source, and it is the right tool for those jobs.
Use GetScreenshot when
The agent only has to see a page, and there is no browser where it runs: a cloud agent, a sandbox, a CI job. Or when you want small images with a known token cost and no Chromium to manage.
| GetScreenshot | Playwright MCP | |
|---|---|---|
| What it is | Hosted screenshot API with an MCP server, a CLI and an agent skill | MCP server that automates a browser with Playwright |
| Where the browser runs | On GetScreenshot's servers | On your machine, or in a container you run |
| Cost | Paid. Plans from $5/month for 2,500 screenshots. No free tier. | Free and open source (Apache-2.0) |
| What the model receives | JPEG images downsized and sliced for vision models, an estimated token cost, and a URL to the original | Structured accessibility snapshots of the page by default. Screenshots through browser_take_screenshot. |
| Clicking, typing, navigating | No. One optional click selector before a single capture. |
Yes. Tools for navigation, clicks, typing, forms, tabs and more. |
| Logged-in sessions | No | Yes. A persistent browser profile, a storage state file, or a browser extension that connects to your existing tabs. |
| localhost and private networks | No. Use render_html, a public preview URL or a tunnel. |
Yes. The browser runs on your machine and can open whatever your machine can. |
| Needs a browser where the agent runs | No. One HTTPS call. | Yes. Node.js 20 or newer and a browser Playwright can launch. |
| Full-page screenshots | Yes, cut into slices of up to 1280x952 so each stays readable | Yes, with the fullPage option |
| Device emulation | 18 named presets through device, or any width and height, per capture |
A --device flag when the server starts, and a tool to resize the window |
| Visual diff | compare_screenshots returns the changed-pixel percentage, a bounding box and a diff image |
No diff tool in its README tool list |
capture_pdf, A4, hosted link |
browser_pdf_save, enabled with --caps=pdf |
|
| Where screenshots end up | Inline in the tool result, and at a public URL for 30 days | In a file on your machine. Image responses to the client are on by default and can be turned off with --image-responses. |
| Install | npx -y getscreenshot-mcp with an API key, or the hosted endpoint |
npx @playwright/mcp@latest |
Playwright MCP details are taken from its README as of November 2026. Check the repository for the current state.
What Playwright MCP Does Well
Things a hosted screenshot API cannot do
It drives the page
The agent can navigate, click, type, fill forms and switch tabs, one step at a time, and read the page after each step.
It keeps you logged in
By default it uses a persistent browser profile, so a login survives between steps. It can also load a saved storage state or connect to tabs you already have open.
It reaches localhost
The browser is on your machine. It can open your dev server, an intranet page or anything behind your VPN.
It reads structure
It works from the page's accessibility tree, so the model does not need vision to act. It also exposes console messages and network requests.
Where GetScreenshot Fits
Jobs where the agent needs to look, and a local browser is unavailable or more than the job needs
No local browser
Cloud coding agents, sandboxes and CI runners often can't launch Chromium. GetScreenshot is an HTTPS call, so it works from any of them.
Model-sized images, known cost
A 1280x800 viewport is one image of about 1,334 tokens. A full page arrives as slices of about 1,564 tokens each, four by default. Every result states its estimate.
Consistent rendering
Every capture goes through the same hosted renderer. The result does not depend on which browser, window size or fonts a developer's laptop or a CI image happens to have.
Device presets per capture
Pass device: "iphone12" on one call and device: "ipad" on the next. Phone presets also send a mobile user agent.
Many captures, no Chromium to manage
Each capture is one request against a monthly quota. There is no browser process to install, update or keep alive.
A link you can share
Every capture has a hosted URL that works for 30 days. The agent can paste it into a pull request or a report. The URL is public, so anyone who has it can open it.
Use Both
They are separate MCP servers and run side by side
{
"mcpServers": {
"playwright": {
"command": "npx",
"args": ["@playwright/mcp@latest"]
},
"getscreenshot": {
"command": "npx",
"args": ["-y", "getscreenshot-mcp"],
"env": { "GETSCREENSHOT_API_KEY": "YOUR_API_KEY" }
}
}
}This is the generic JSON shape. Your client may use a different file or format. See the client setup pages.
-
On your laptop: Playwright MCP for the dev server on
localhost, for flows behind a login and for debugging an interaction. - In a cloud agent or CI: GetScreenshot for a look at the preview deployment, where there is no browser to drive.
-
For a final visual check:
compare_screenshotson production against the preview, to see what changed and where. -
For HTML that is not deployed:
render_htmlon a component or email the agent just wrote.
A note on context size. Playwright's own README points coding agents to the Playwright CLI as the more token-efficient option, because CLI calls avoid loading tool schemas into context. GetScreenshot has the same choice: an MCP server, or a CLI and agent skill.
When Playwright MCP Is the Better Choice
Pick Playwright MCP, and skip GetScreenshot, if:
-
The pages you need are on
localhostor a private network and you cannot expose them - The agent has to log in, or take several steps to reach the state you want to see
- You want the agent to act on the page, and a picture of it is not enough
- The pages show data that must not be stored at a public URL
- You already have a browser on the machine and do not want a paid service
GetScreenshot is the better fit when the agent runs somewhere without a browser, when image size and token cost matter, or when all it needs is to see a public page.
Add Screenshots to Your Agent
Plans start at $5/month for 2,500 screenshots. The MCP server, CLI and agent skill work on every plan.