MCP integration
Take website screenshots with AI agents, over MCP.
One server URL gives an MCP-compatible assistant the whole product: capture a page, schedule a recurring watch, act on the page first, capture from another country, and read back what changed next week.
- Instant captures and recurring scheduled ones, both as tools.
- Turn raw HTML the agent already has into an image.
- Multi-step actions: log in, fill forms, click, then capture.
- Capture from 150+ locations, and run domain research across hundreds of domains.
25 free screenshots / month · no credit card · no phone number
"Watch their pricing page every hour and tell me when it changes."
create_screenshot_config({
url: "competitor.com/pricing",
scheduled_screenshot: true,
scheduled_every: 1,
scheduled_interval: "hours",
multi_step_actions: [
{ type: "change", selector: ".plan-price",
send_to: "webhook",
url: "https://yourapp.com/hooks/pagepixels" }
]
})
→ screenshot_config_id: 827b7f5d-da55-44aa
→ first capture runs immediately
How to connect the PagePixels MCP server
No local process to run, nothing to install, and no API key pasted into a config file. It is a remote server your client authorises.
-
Create a free account
Sign up for PagePixels. The free plan carries every tool on this page, so there is nothing to upgrade before an agent can use it.
25 screenshots / mo -
Add the server URL
Paste the server address into your MCP client's server list, the same way you would add any other remote MCP server.
https://mcp.pagepixels.com/mcp -
Authorise the connection
Approve access when the client prompts. OAuth 2.1 handles it, so no API secret ends up sitting in a config file on disk.
oauth 2.1 -
Ask for a screenshot
The client lists the tools automatically. Ask in plain language and the agent picks the right one, whether that is a one-off capture or a standing watch.
"capture their pricing page"
Works with any MCP-compatible assistant or your own agent framework. The server's repository is at github.com/PagePixels/mcp-server, and the tool list, arguments, and examples are in the MCP guide.
The whole API, exposed as tools an agent can choose between.
Not one render tool. Sixteen, in five groups, so an agent can capture, schedule, watch, relocate, and analyse without you writing glue.
-
Capture now
instantOne-off captures of a live URL, or of HTML the agent is already holding, with every rendering option the API takes.
-
snap -
snap_html -
capture_screenshot
-
-
Schedule and manage
recurringCreate a recurring watch, read it back, change its interval, list what is already running, and delete what is finished.
-
create_screenshot_config -
get_screenshot_config -
update_screenshot_config -
list_screenshot_configs -
delete_screenshot_config
-
-
Read what changed
watchingPull the change notifications a watch has produced since the agent last looked, with the values on either side of the move.
-
list_config_change_notifications -
list_all_change_notifications
-
-
Choose a location
150+ placesList the available countries, states, and cities, then capture through one so a geo-gated page renders the way a local sees it.
-
list_real_locations
-
-
Analyse and extract
understandingAsk a question of an image and get a written answer, or run structured extraction across a list of domains and collect the report.
-
analyze_any_image_with_ai -
create_domain_research -
get_domain_research_status -
get_domain_research_report -
list_domain_research_reports
-
Arguments, defaults, and worked examples for every tool are in the MCP guide.
Your agent can already see the web. Give it a memory of it.
Most browsing tools are stateless. The agent fetches a page, reasons about it, and the session ends. Ask the same agent next month what that page used to say and it has nothing, because nothing was kept.
These tools write to an account instead of a context window. A config the agent creates today keeps capturing after the conversation ends, every capture is dated and kept in the archive, and change notifications queue up until something asks for them.
That turns three separate jobs into one: the agent decides what to watch, the platform does the watching, and the agent reads the result whenever it next runs. No cron on your side, no browser to babysit, and no scraper to keep alive.
- Fetches a page inside one session Creates a watch that keeps running after it
- Remembers nothing once context clears Every capture dated and kept in the archive
- Cannot tell you what changed Change notifications queue until something reads them
- Sees the page as the datacentre sees it Captures through 150+ real locations
- Stops at pages that need no login Logs in, fills forms, and clicks before capturing
An agent can act on the page before it captures it.
Multi-step actions are available through the tools, not only in the web app. So the agent can log in with stored credentials, dismiss the consent wall, click into the tab that matters, wait for the element to render, and only then capture.
It also works the other way round. When the thing worth keeping is not a URL at all, the agent hands over raw HTML and gets an image back, which is how a rendered report, an email, or a JSON payload becomes something a person can glance at.
Captures can run through a Real Location in 150+ countries, states, and cities, so an agent checking a geo-gated page sees what a customer in that market sees.
-
Agent creates the config create_screenshot_config
-
Fill email and password text_field ×2
-
Click submit click
-
Wait for the dashboard wait_for_selector
-
Capture on an interval hourly · Germany
-
Agent reads the changes later list_all_change_notifications
Four jobs worth handing to an agent
Each one is a tool call or two, and the work continues after the conversation is over.
-
competitive research
Watch a rival while you sleep
The agent sets up hourly captures of a pricing page with a watched selector, then reports the differences the next time you ask, rather than re-reading the page from scratch.
create config · read notifications -
reporting
Turn data into something visual
The agent renders the HTML it has just assembled into an image and posts it where the team already looks, without a headless browser in your stack.
snap_html · send to Slack -
market checks
Same page, five countries
Capture a landing page from each market it runs in, so the agent can compare what customers in each region are actually shown.
snap · 5 Real Locations -
research at scale
Structured facts from many sites
Point domain research at a list of companies and collect the same fields for each, as JSON the agent can reason over.
create_domain_research · JSON
The MCP server is not an enterprise feature here.
This list is the same on the free plan and the largest one. The only number that changes between plans is how many screenshots you take.
Capture
5 INCLUDED- Multi-step capture Log in, click, fill forms, and wait for elements before the shutter fires.
- Scheduled screenshots Recurring captures from every five minutes to once a year.
- Real Locations Capture through a proxy network in 150+ countries, US states, and cities.
- Ad, tracker, and banner removal Strip cookie consent walls, ads, and third-party scripts before capture.
- CSS and JS injection Restyle or manipulate the page before the capture is taken.
Understand
4 INCLUDED- AI analysis Ask a question of any capture and get a written answer back with the image.
- Custom HTML screenshots Render email, private spreadsheet rows, or API payloads into an image.
- HTML extraction Pull fully rendered HTML from JavaScript-built pages, whole page or one selector.
- Website domain research Extract structured fields across a list of domains, as JSON or CSV.
Watch
3 INCLUDED- Change notifications Watch a selector and get a webhook, Slack message, or Zap when it moves.
- Archived history Every capture is kept and listable through the API, with dated thumbnails.
- CDN embed URLs A permanent link that always shows the latest capture, no API key exposed.
Deliver
3 INCLUDED- Every integration Zapier, Make, n8n, Power Automate, Slack, Dropbox, webhooks, Chrome, and MCP.
- Full REST API Every option is a parameter, with NPM, PyPI, and RubyGem packages.
- Custom headers and cookies Send your own headers, cookies, user agent, language, and time zone.
Screenshot MCP, answered.
What is the PagePixels MCP server URL?
https://mcp.pagepixels.com/mcp. Add it to your MCP client's server list as a remote server, approve the OAuth prompt, and the tools appear. There is nothing to install and no local process to keep running.
Which AI assistants work with it?
Any MCP-compatible client, and your own agent framework if it speaks the protocol. Because it is a remote Streamable HTTP server rather than a local stdio binary, the client only needs the URL and permission, not a machine to run a process on.
How does authentication work?
OAuth 2.1, handled between your client and the server, so no API secret is pasted into a config file on disk. Revoke access from your account and every connected client loses it at once.
Is MCP included on the free plan?
Yes. The MCP server is on every plan including the free one, with the same tools and the same options. The free plan is 25 screenshots a month with no credit card and no phone number, which is enough to build an agent workflow against before paying anything.
Can an agent set up a recurring watch, not just take one screenshot?
Yes, and this is the part most screenshot tools cannot do. The agent calls create_screenshot_config with an interval and, optionally, a watched selector. That config keeps capturing after the conversation ends, and the agent can list, read, update, or delete it later.
How does the agent find out what changed?
It calls list_all_change_notifications, or list_config_change_notifications for a single watch, and gets back what has been recorded since it last looked, including the selector that moved and the values on either side. So an agent that runs once a week can still report on a page captured hourly.
Does agent use count against my screenshots?
Captures do: one capture is one screenshot, whoever asked for it. AI analysis and domain research run on their own meters. Listing configs, reading change notifications, and looking up locations cost nothing on their own.
