Automate browser tasks by reading, clicking and filling forms
The poster uses Claude in Chrome to read page content, click buttons and fill in forms, and asks about alternatives for browser automation.
Done withClaude in Chrome
- What they did
- The poster says they use Claude through a Chrome extension to automate browser work by reading page content, clicking buttons, and filling forms, then asks what other setups people use for the same workflow. In the replies, people point them to alternatives like chrome-devtools MCP, Browser Use, Stagehand, Playwright, Browserbase-style managed browsers, and Anthropic’s computer-use API, with several commenters stressing screenshots, logs, retries, and human approval for risky steps.
- How it went
- There is no reported result, metric, or finished project in the thread; it is mainly a request for alternatives and a comparison of tools. The clearest limitations mentioned are flaky behavior, no multi-tab parallel work in one MCP setup, bot detection, slowness, and cost.
- Worth knowing
- The most practical catch commenters repeat is reliability: use human approval for money or account actions, and confirm page state with screenshots before the agent acts.
Try it yourself with Claude in Chrome
On [website URL], read the page and pull out [the information I need, e.g. a list of items with prices]. Then fill in the form on [page or URL] using these details: [field values], and click through to the review step. Stop before the final submit button and show me everything you entered so I can confirm.
Want help getting this working? Join the Vibe Coding Academy ↗ Run by Alex Finn, who also runs this site.
Source: Reddit · May 15, 2026
Related use cases
Build a balloon-camel simulator for an LED Burning Man totem
Author used ChatGPT for reference images and Claude Code to build a Go balloon physics simulator with a web frontend, with Claude looping on screenshots and camera control.
Book flights and hotels, stopping at payment for approval
Claude Cowork drove the browser through Claude in Chrome to book 8 flights and 5 hotels, filling traveler details and seats and pausing at payment for approval.
Audit and fix a Home Assistant setup
The writer had Claude audit their Home Assistant config, using the Chrome connector to click through the dashboard and paste config blocks into editors when the MCP couldn't edit directly.
Run a solo company with eight agent departments
A founder set up CEO, CFO, COO, lawyer, accountant, marketing, CTO, and improver agents that share memory and consult each other.
Process supplier invoices from Gmail through approval and payment
An n8n pipeline extracts invoice data, checks duplicates and suppliers, routes high-value invoices for approval, then creates accounting entries and schedules payment.
Run routine store checks on delayed orders and stock
A commenter runs Claude against their store to check delayed orders, stalled tracking, and stock issues; the poster is connecting an assistant via MCP.
Get 5 like this every morning
The best things people got an AI agent to do, each with the prompt to try it.