Set up Cloak Biz Scraper for your AI

Complete this guide once. At the end, Claude or ChatGPT can launch and operate a cloaked browser through MCP, using infrastructure and accounts you control.

You can then choose what the browser does:

• build a scheduled business-listing watch;
• open a public page that blocks an ordinary AI browser; or
• give the agent another browsing task with the prompt pattern at the end of this page.

flowchart LR
    A[Deploy Cloak Biz Scraper] --> B[Add the emailed<br>CloakBrowser Pro key]
    B --> C[Add a residential proxy]
    C --> D[Connect the /mcp URL<br>to Claude or ChatGPT]
    D --> E[AI calls create_instance<br>and agent_browser]

Before you start

You need:

1. A Railway account for hosting.
2. A paid CloakBrowser licence. The Pro key is sent by email after checkout; it does not come from a browser manager.
3. A residential proxy account. This guide uses Evomi Core Residential.
4. Claude or ChatGPT with support for a custom remote MCP connection.

The project is open source, but hosting, proxy traffic, CloakBrowser, and AI usage may cost money. Check each provider's current plan before subscribing.

1. Deploy the project on Railway

1. Open the project's one-click Railway template.
2. Deploy the template and wait for the service to become healthy.
3. In Railway, open Settings → Deploy → Serverless, enable it, then redeploy. The new setting applies to the newly deployed container.
4. Open Variables and copy APP_SECRET. Treat it as a password.
5. Open the public Railway URL and sign in with APP_SECRET.

Railway creates a persistent /data volume for settings, browser profiles, and recent task history. Never put APP_SECRET, the browser key, or proxy credentials in an AI prompt, issue, or screenshot.

Watch the deployment walkthrough

This video shows the Railway deployment and first launch. It does not configure the residential proxy or Notion. Complete the proxy setup in Step 3 below before testing protected sites. If you are building the daily listing workflow, its Notion setup comes afterward.

2. Get and add the CloakBrowser Pro key

1. Open the official CloakBrowser website and find its pricing.
2. Choose a plan with enough concurrent browser sessions for your expected workload and complete checkout with an email address you can access.
3. Follow the instructions in the purchase email to obtain the licence key. Check spam and the address on the receipt if the message does not arrive.
4. In Cloak Biz Scraper, open Settings → Browser licence.
5. Paste the emailed key. Leave the browser version blank to use the current build, then select Save & verify.

The verification must report a working Pro build before you continue. The key is supplied by CloakBrowser's checkout email flow; the CloakBrowser Manager app is not part of this setup.

3. Add the Evomi residential proxy

Generate the credentials

1. Create an Evomi account and activate Core Residential traffic.
2. Sign in at my.evomi.com. Under My Products, open Core Residential, then select Proxy Generator.

1. Under Format Settings, choose HTTP and the host-first format hostname:port:username:password. Start with Worldwide; the scraper can add its own country and region selection.

1. Select Evomi's copy button. It copies one complete formatted string, not a separate username and password. Split the copied value into the four scraper fields:

core-residential.evomi.com:1000:YOUR_USERNAME:YOUR_PASSWORD
| proxy host               |port| username   | password

Remove an initial http:// if Evomi included it. The first segment is Proxy host, the second is Proxy port, the third is Proxy username, and everything after the third colon is Proxy password.

Save and test it

1. In Cloak Biz Scraper, open Settings → Evomi Proxy.
2. Enter the host, port, username, and password in their separate fields. The current Core Residential endpoint uses host core-residential.evomi.com and HTTP port 1000; use the values shown in your Evomi dashboard if they change.
3. Choose the country and optional region where the browser should appear.
4. Select Save & test. Continue only when the test shows that traffic is leaving through the residential proxy.

If the proxy is incomplete or rejected, the scraper fails visibly instead of silently using the Railway server's datacenter IP. If credentials are exposed, use Reset Proxy Key in Evomi and replace them in the scraper.

4. Connect the scraper MCP to your AI

Open Cloak Biz Scraper's Connect tab. Copy the public MCP URL, which ends in /mcp:

The screenshot shows a local development address. Your AI needs the public HTTPS URL of your Railway deployment:

https://your-server.up.railway.app/mcp

The connection uses OAuth. When Claude or ChatGPT redirects you to your own scraper, enter APP_SECRET on that scraper authorization page and approve the connection. Do not paste the secret into a chat prompt.

Claude

1. Open Customize → Connectors.
2. Select Add → Add custom connector.
3. Name it Cloak Biz Scraper, paste the public /mcp URL into Remote MCP server URL, and continue.
4. Complete authorization on your scraper's page.

ChatGPT Work

ChatGPT Work is available on individual paid accounts as well as managed workspaces, subject to rollout. In the current individual-account interface:

1. Open Plugins → Create app.
2. Enter Cloak Biz Scraper, select Server URL, paste the public /mcp URL, and keep Authentication set to OAuth.
3. Read the custom-server warning. Acknowledge it only when the URL is your own deployment, then create and authorize the app.

Custom MCP action availability varies by ChatGPT plan and product rollout. A connection is not proven until the tool calls in the next step work. If the AI can read server_info but cannot call action tools, that account cannot yet use the raw custom-MCP workflow; use a plan with full MCP actions or Claude.

5. Run a harmless connection test

Paste this prompt into the AI that has the new connection:

Use my connected Cloak Biz Scraper MCP for this test. Do not use your built-in browser.

1. Call server_info and confirm that the server reports a verified Pro browser and a
   working residential proxy. Do not reveal any credential values.
2. Call create_instance with profile="Default" and geoip=true. Save the instance_id.
3. Call agent_browser with that instance_id and command="navigate https://example.com".
4. Call agent_browser again with command="get title", then with command="get url".
5. Report the title and final URL.
6. Call close_instance for only the instance you created.

If a tool is missing or a call fails, report the exact tool and error. Do not substitute
ordinary web browsing and do not claim the test passed.

The expected title is Example Domain. During the run, the browser appears on the scraper's Overview and Browsers tabs.

The most useful connected tools are:

Job	MCP tool
Check configuration and capacity	server_info
Launch a browser	create_instance
Navigate, read, click, fill, or take a screenshot	agent_browser
Inspect or refresh links for open browsers	list_instances, get_instance
Close a browser	close_instance
Give a profile a new exit IP on its next launch	new_proxy_session
Sweep supported listing-search pages	scrape_listings, then get_scrape_listing_results
Save a page's readable content into an existing Notion page	archive_page

6. Give the AI an explicit browsing prompt

For any browsing task, name the connection and the two tools. This prevents the agent from retrying its ordinary browser:

Use the Cloak Biz Scraper MCP, not your built-in browser, to open [URL] and [GOAL].

Call create_instance(profile="Default", geoip=true), save its instance_id, and use
agent_browser with that instance. Navigate to the URL, inspect it with snapshot -i or read,
and take a fresh snapshot after each page change. Report the final URL and requested facts.
Close only the instance you created when you are done. If the page needs a login, CAPTCHA,
payment, or consent from me, stop and ask me to take control.

Next, read Browse a page that blocks ordinary AI browsers for more prompt examples. Use Monitor and take control to watch a run or finish a human-only step, and Advanced controls to tune proxy sessions, capacity, profiles, and disk usage.

If the test fails

• No MCP tools appear: reconnect the server so the client refreshes its cached tool list.
• The MCP URL cannot be reached: use the Railway public HTTPS URL, not 127.0.0.1.
• OAuth returns to the scraper: that is expected; enter APP_SECRET only on your own scraper's page.
• The browser licence fails: re-copy the key from CloakBrowser's email and use Save & verify.
• The proxy test fails: regenerate or reset the Evomi proxy key, split the copied string into four fields, and use Save & test again.
• The site still blocks the browser: inspect the saved evidence, wait before one bounded retry, and see the protected-site and advanced-controls guides. No browser or proxy can guarantee access to every page.

This setup was checked against the repository's live MCP implementation and the provider flows listed in the verification record.