Browse a page that blocks ordinary AI browsers¶

Use this guide when an AI's normal web tool cannot open a public page because the site blocks datacenter traffic or browser automation.

Cloak Biz Scraper gives the AI a patched Chromium browser behind your residential proxy. The AI must call the scraper's MCP tools explicitly. “Try the website again” is often too vague; the agent may retry the same blocked browser.

Before you start¶

Complete Set up Cloak Biz Scraper for your AI first. That guide covers the Railway deployment, emailed CloakBrowser Pro key, residential proxy, MCP connection, and a harmless end-to-end test.

Do not continue until your agent can call these tools from the Cloak Biz Scraper MCP:

server_info
create_instance
agent_browser
get_instance and list_instances
close_instance

This guide does not require the business-listing workflow or Notion. Add Notion only when you want to archive a page there.

flowchart LR
    A[Ordinary browser is blocked] --> B[create_instance<br>profile: Default]
    B --> C[agent_browser<br>navigate URL]
    C --> D[agent_browser<br>snapshot or read]
    D --> E{Need another action?}
    E -->|Yes| F[click / fill / press<br>one command at a time]
    F --> D
    E -->|No| G[Summarize or extract]

The shortest useful prompt¶

Replace the URL and the information you want:

Use the Cloak Biz Scraper MCP, not your built-in browser, to open
https://example.com/page and tell me [what you need].
Read the connected scraper's tool instructions and follow them.

The connected MCP tool descriptions explain how to launch, operate, and close the browser. The longer prompt below is useful only when an agent still substitutes its ordinary browser.

When the ordinary browser already failed¶

The ordinary browser failed because the site blocked automated or datacenter traffic.
Do not retry that browser.

Use the Cloak Biz Scraper MCP:
1. Call create_instance(profile="Default", geoip=true).
2. Save the returned instance_id.
3. Call agent_browser(instance_id=<id>, command="navigate <URL>").
4. Call agent_browser with command="snapshot -i" or command="read" to inspect the page.
5. Use one agent_browser action per call. Take a fresh snapshot after navigation, clicks,
   form submissions, or other page changes.
6. Return the requested facts with the page URL. If the page still fails, report the actual
   error instead of claiming you read it.

Useful `agent_browser` commands¶

The command string is one browser action. The agent calls the MCP tool again for the next action.

Goal	Command example
Open a page	`navigate https://example.com/page`
Inspect interactive elements	`snapshot -i`
Inspect the page with URLs	`snapshot -i -u`
Read the visible page	`read`
Click a snapshot reference	`click @e3`
Fill a field	`fill @e3 "California"`
Submit a form	`press Enter`
Confirm the final address	`get url`
Confirm the page title	`get title`
Read one element	`get text @e3`
Go back	`back`
Reload once	`reload`
Capture the visible page	`screenshot`
Capture the full page	`screenshot --full`

Snapshot references such as @e3 are temporary. The agent should take a new snapshot after the page changes instead of guessing an old reference.

A prompt for a search or multi-step page¶

Use the Cloak Biz Scraper MCP for this browsing task.

Call create_instance(profile="Default", geoip=true), then use agent_browser with its
instance_id. Navigate to [START URL]. Take snapshot -i, fill or click the visible controls
needed to [GOAL], and take a fresh snapshot after every page change. Use get url to capture
the final filtered URL and return it to me with [THE FACTS I NEED].

Do not use the built-in web browser. Do not guess hidden controls or reuse stale element
references. Stop and tell me if the site requires a login, CAPTCHA, payment, or acceptance
of terms that needs me.

This is also the safest way to create a BizBuySell seed URL: have the cloaked browser apply the filters, then read the final address with get url.

Optional custom instructions¶

If the agent repeatedly retries its ordinary browser, add this to its personal or project instructions:

When I ask you to open a public website and ordinary web access is blocked by anti-bot or
datacenter-IP detection, use my connected Cloak Biz Scraper MCP.

Call create_instance with the durable profile "Default", then drive that instance with the
agent_browser tool. Use navigate, snapshot -i or read, and one action per tool call. Take a
fresh snapshot after page changes. Do not claim a page was read unless agent_browser returned
its content. Ask me to take over for login, CAPTCHA, payment, consent, or other human-only
steps. Close only instances you created for the current task.

Keep this instruction scoped to access failures. A normal browser is faster for pages that already work, and unnecessary protected-browser traffic consumes proxy bandwidth.

Save the page to Notion when you need a record¶

Browsing and archiving are different operations.

This optional section requires a Notion internal integration configured in the scraper with access to the destination page, plus the AI's Notion connector to read it afterward. Skip it if you only need to browse and return facts in the chat.

If an existing Notion page should hold a permanent copy, tell the agent:

Use the Cloak Biz Scraper MCP tool archive_page with url=<PAGE URL> and
notion_page_id=<EXISTING NOTION PAGE ID>. Wait for it to finish. Then use the Notion MCP to
read the archived page body and confirm a Source Content section was appended.

archive_page takes roughly a minute, appends to the page, and does not edit its properties. Do not call it twice just because it is slow: another successful call appends another copy.

Logins, CAPTCHAs, and blocked IPs¶

The Browsers tab lets you watch the browser and take control when a site needs a human:

A live cloaked browser with residential proxy status

The Default profile is durable. Its cookies and logins survive a browser relaunch, so use the same profile for continuity.
Do not put passwords or one-time codes in prompts. Open the scraper's Browsers tab and take control of the live browser when a site needs you to sign in.
CloakBrowser does not solve CAPTCHAs. Complete legitimate verification yourself.
If a residential exit IP is blocked, close the current instance, call new_proxy_session(name="Default"), then create a new instance. Do this once and report the result; do not rotate in an aggressive loop.
A browser closes after 15 minutes idle or 60 minutes total. If a CDP link expires, call get_instance or list_instances for a fresh link rather than launching duplicate browsers.

See Monitor a browser and take control for the complete live-preview, takeover, task-history, and evidence workflow. See Advanced controls before changing proxy sessions, browser capacity, profiles, or stored data.

Use the browser only for public pages or sites you are authorized to access. Respect site terms, access controls, rate limits, and paywalls.

Verify the setup once¶

Ask the agent to perform a harmless read-only check:

Use Cloak Biz Scraper to open https://example.com with profile="Default". Report the page
title and final URL, then close the instance you created. Also report whether the server says
the browser is using the configured residential proxy. Do not submit forms or save data.

Then repeat the check on a public page that previously blocked the ordinary browser. If it still fails, open Cloak Biz Scraper Settings and retest the browser licence and proxy before changing prompts. Use the bounded proxy-rotation procedure in Advanced controls only when the current residential exit is clearly blocked.