TL;DR
- Claude Code with Playwright MCP can navigate any web interface by description—no API required
- One homeowner reduced monthly web interface tasks from 8-10 hours to 1-2 hours with automated browser control
- Key capability: Claude reads page elements, clicks buttons, fills forms, and completes multi-step workflows autonomously
- Best for: Router config, utility portals, subscription management, smart home dashboards, and dark pattern navigation
- Critical practice: Use screenshots as checkpoints before sensitive actions, and implement proper credential management
Claude Code with Playwright MCP transforms tedious web interface navigation into simple descriptions—clicking through forms, menus, and dark patterns autonomously.
Kevin’s home network was a mess.
Router settings that needed adjusting. Smart home dashboards that required configuration. Utility portals that demanded monthly logins. Subscription services that buried the cancel button three screens deep.
Every task required the same tedious process: open browser, navigate, login, click through menus, find the setting, change it, save, confirm.
“I spent hours each month just clicking through web interfaces. The same interfaces. The same sequences. Over and over.”
He knew automation existed for APIs. But most of these services didn’t have APIs. They had web interfaces designed for humans. Slow, clicking humans.
The Playwright Discovery
Kevin learned about browser automation tools. Playwright could control a browser programmatically — navigating pages, clicking buttons, filling forms, reading content.
More importantly, someone had built an MCP server for Playwright.
“That meant Claude could control my browser. I could describe what I wanted in plain English, and Claude would figure out the clicks.”
The combination was powerful: Claude’s understanding plus Playwright’s hands.
The Router Experiment
Kevin started with his router.
The router’s admin interface was a nightmare. Fifteen tabs, nested menus, settings scattered across multiple pages. Changing the Wi-Fi password required six screens.
“I said to Claude: ‘Log into my router at 192.168.1.1, navigate to wireless settings, and change the Wi-Fi password to [new password]. Then update the guest network password to [different password].’”
Claude launched a browser. Navigated to the router IP. Encountered the login screen. Entered credentials Kevin had provided. Found the wireless settings by reading page elements. Located the password fields. Updated both passwords. Saved changes.
“It took Claude about three minutes. It took me fifteen when I did it manually. And Claude could do it again instantly.”
The Expanding Scope
Kevin found more targets.
Internet provider portal: Monthly usage wasn’t available via API. But the web portal showed it. Claude could log in, navigate to the usage page, extract the numbers, report back.
Smart home dashboard: The thermostat’s schedule lived in a web interface. Claude could navigate there, read the current schedule, suggest changes based on Kevin’s preferences, implement them.
Utility bill payment: The electric company’s autopay settings were buried in account preferences. Claude found them, verified the payment method, confirmed the schedule.
Subscription management: Several services made cancellation deliberately difficult. Claude navigated the dark patterns, found the cancel buttons, completed the process.
“Any web interface I used regularly became automatable. I just had to describe what I wanted done.”
The Smart Home Integration
Kevin’s home had multiple smart systems. Different vendors. Different apps. Different web interfaces.
Unifying them through APIs would have required development work. Unifying them through browser automation required only descriptions.
“Turn off all smart devices before we leave for vacation.”
Claude logged into each system’s web interface — the lights, the thermostat, the security cameras — and made the changes across all of them.
“I didn’t need an integration layer. I had Claude navigating each interface like I would, just faster and without mistakes.”
The Shopping Automation
Kevin expanded to commerce tasks.
Price monitoring: “Check the price of [product] on Amazon and these three other sites. Tell me who has the best deal.”
Claude visited each site, searched for the product, extracted prices, compared them.
Reorder routine items: “I need more paper towels. Order from whichever site has the best price under $30 including shipping.”
Claude checked sites, found options, compared total costs, completed the purchase on the winner.
“I wasn’t just reading prices — Claude was actually checking out. The whole loop, automated.”
The Form-Filling Capability
Web forms were everywhere. And they were tedious.
Kevin discovered Claude could fill forms intelligently.
Insurance quote requests: “Fill out this insurance quote form using my standard information. Stop at payment.”
Claude navigated multi-page forms, selecting options, entering text, clicking through to the quote summary.
Government applications: “Complete this permit application using the project details I gave you. Show me the summary before submitting.”
Claude handled complex forms with conditional logic — showing certain fields based on previous answers, navigating branching paths.
“Every form became a conversation instead of a chore.”
The Screenshot Verification
Kevin learned to trust but verify.
Before Claude completed sensitive actions, he asked for screenshots.
“Navigate to my bank’s bill pay section. Take a screenshot before making any changes.”
Claude captured the state. Kevin reviewed. Then he approved the next step.
“For anything involving money or permanent changes, I had Claude show me what it was about to do. Screenshots gave me checkpoints.”
The Monitoring Setup
Beyond one-time tasks, Kevin set up monitoring.
“Every morning, check these three sites for any notifications or alerts. Summarize what needs attention.”
Claude ran the routine. Logged into his insurance portal (any messages?). Checked his utility account (any issues?). Reviewed his investment dashboard (any alerts?).
“I got a daily briefing of everything that needed my attention across all my web accounts. Without visiting any of them myself.”
The Dark Pattern Navigation
Some sites were designed to frustrate users. Hidden settings. Confusing flows. Deliberate friction.
Claude was immune to dark patterns.
“Find the ‘delete my account’ option on [service]. They hide it well.”
Claude systematically explored menus, searched help pages, found the buried option.
“Cancel my subscription to [service]. They probably have multiple confirmation screens.”
Claude clicked through whatever obstacles existed. No frustration. No giving up.
“Dark patterns work on humans because we get tired or confused. Claude just keeps navigating.”
The Security Considerations
Browser automation with login credentials required care.
Kevin implemented safeguards:
Credential management: Passwords stored in a password manager. Claude retrieved them via MCP integration rather than having them in plaintext prompts.
Session isolation: Automation ran in a separate browser profile. No cookies shared with his main browsing.
Audit logging: Every automated session captured screenshots at key steps. Kevin could review what Claude had done.
Scope limits: Certain sites (banking, healthcare) required explicit approval before Claude could navigate them.
“I trusted Claude with my router. I verified more carefully with my finances.”
The Time Recapture
Kevin estimated his time savings.
Before automation: 8-10 hours monthly on web interface tasks After automation: 1-2 hours monthly (mostly reviewing what Claude did)
“The savings weren’t dramatic for any single task. But aggregated across all the web interfaces in my life, it was significant.”
Those hours went to things he actually wanted to do.
The Limitation Recognition
Not every site worked perfectly.
Some had anti-automation protections. CAPTCHAs blocked progress. Bot detection triggered lockouts.
“I couldn’t automate everything. Some sites really didn’t want automation and were good at blocking it.”
For those sites, Kevin still did things manually. But they were the minority.
The Expansion Ideas
Kevin saw more possibilities.
Work automation: Internal tools with web interfaces. Status updates. Time tracking. Report generation.
Research tasks: Competitor websites. Price monitoring across industries. Content aggregation from multiple sources.
Testing: His own projects had web interfaces. Claude could test them as a user would.
“Anywhere there’s a web interface, there’s an automation opportunity. The question is whether it’s worth setting up.”
The Philosophical Shift
Kevin’s relationship with web interfaces changed.
“I used to dread anything that required logging into a web portal. Now it’s just a description of what I want done.”
The browser became a tool Claude operated rather than something Kevin navigated. The interfaces stopped mattering — only the outcomes mattered.
The Recommendation
For others exploring browser automation:
“Start with repetitive tasks on low-risk sites. Your router. A newsletter subscription. Something where mistakes don’t matter.”
“Build up to sensitive tasks gradually. Understand how Claude handles errors before trusting it with money.”
“Use screenshots as checkpoints. Browser automation can go wrong. Visual verification catches mistakes.”
“Respect sites that don’t want automation. If they block you, accept it. Don’t try to circumvent legitimate protections.”
