Agent Browser (Browserbase)
Agent Browser connects your Claw to a real, cloud-hosted browser via Browserbase. Your Claw can navigate websites, click buttons, fill forms, take screenshots, and extract data from any page — including those that require login or block scraping.
The problem
Web search returns links. Agent Browser returns content. When you need your Claw to read a live pricing page, submit a form, log into a dashboard, or interact with a web app, web search isn't enough. Agent Browser gives your Claw full browser control — it sees exactly what a human would see.
What you need
- Agent Browser skill installed
- A Browserbase account and API key (browserbase.com (opens in a new tab))
Install the skill:
npx clawhub@latest install agent-browserSteps
Create a Browserbase account
Sign up at browserbase.com (opens in a new tab). The free tier is sufficient for most personal workflows. Grab your API key from the dashboard.
Configure your API key
openclaw env set BROWSERBASE_API_KEY=your_key_hereYour Claw will pick this up automatically in any session, including isolated cron runs (as long as TOOLS.md notes that the browser is available).
Update TOOLS.md
Add the browser to your integration record:
Update ~/TOOLS.md to note that Agent Browser (Browserbase) is configured and available.Test with a simple fetch
Use Agent Browser to visit https://stripe.com/pricing and tell me the current price for the Starter plan.If the browser is connected correctly, your Claw will navigate to the page and return the actual live content.
Try a more complex task
Use Agent Browser to log into [your tool] at [URL] with these credentials: [username / password saved in memory].
Once logged in, navigate to the billing page and tell me the current plan and renewal date.What Agent Browser can do
| Task | Example |
|---|---|
| Read live web content | "Check the current Anthropic API pricing page" |
| Monitor for changes | "Check [URL] daily and notify me if the pricing changes" |
| Extract structured data | "Visit [job board URL] and extract all senior engineer listings into a table" |
| Interact with forms | "Fill in and submit the contact form at [URL] with these details" |
| Take screenshots | "Screenshot the homepage of [competitor] and save it" |
| Navigate multi-step flows | "Log into [dashboard], navigate to Reports, and download the CSV for last month" |
Patterns
Competitive monitoring (daily cron):
openclaw cron add \
--name "Competitor pricing check" \
--cron "0 9 * * 1" \
--session isolated \
--message "Read ~/TOOLS.md. Use Agent Browser to check [competitor]/pricing. Compare to last week's saved snapshot in memory. If pricing has changed, send me a Telegram message with what changed." \
--announce \
--channel telegramResearch assistant:
I'm evaluating these 5 tools: [list]. Use Agent Browser to visit each one's pricing page and give me a comparison table with tiers and prices.Form submission:
Use Agent Browser to go to [URL], fill in the contact form with my name and email from memory, enter "Interested in enterprise pricing" in the message field, and submit.Tips
Use memory to store login credentials. Don't paste credentials into every message. Save them to QMD memory or reference them from TOOLS.md:
Save my [service] login credentials to memory so Agent Browser can use them.Browserbase handles CAPTCHAs and JavaScript. Unlike raw HTTP scraping, Browserbase runs a real browser — so most CAPTCHAs and JavaScript-rendered pages work out of the box.
Screenshot before and after for verification. For important actions (form submission, data extraction), ask your Claw to take a screenshot as confirmation.
Browserbase uses stealth browsing to avoid bot detection. Most sites that block scraping tools will still work. Sites with aggressive bot protection (some financial dashboards) may still require manual interaction.
Pair Agent Browser with the Indexable Memory (QMD) recipe to build a research workflow that saves everything it browses — so your Claw can retrieve any page it's ever visited without re-browsing it.