QUICK START

Quick start: connect an Agent to Baizhi Cloud Browser

Read the live setup docs, reuse secure configuration, perform read-only checks, then run a real web task, retrieve files, and explicitly end the browser run.

To connect Baizhi Cloud Browser, have your Agent read the live instructions, inspect existing credentials, and check current pricing. Create a browser run only when you have a real web task. A CDP client such as Playwright performs page actions.

01

Read the source of truth

Start with the setup reference and Agent workflow. Reuse securely stored CAPYBROWSER_BASE_URL and CAPYBROWSER_API_KEY; never paste a real key into chat or this website.

02

Check before creating a run

Read /version, validate the existing credential, and read current pricing. These checks do not start a browser run.

03

Give your Agent a real web goal

Specify the page, the action, and the result you need. Your Agent manages one run through MCP or HTTP, waits until ready, obtains a one-time CDP address, and uses its browser client for page work.

04

Step in when needed

If the site requires your sign-in or approval, your Agent sends the human-action link for that run. Complete the step on the visible page and click the return-control action. The Agent reconnects to the same run and checks the page.

05

Retrieve and close

Retrieve and verify downloads before ending the run your Agent created. Confirm the browser is released; check saved state and final billing separately. Closing a CDP connection does not stop the run or billing.

COPY TO YOUR AGENT

A practical starting instruction

Edit it for your task. Copying does not read a key or create a run.

First read https://capybrowser.app.baizhi.cloud/docs/installation.md and https://capybrowser.app.baizhi.cloud/docs/skills.md. Inspect the Baizhi Cloud Browser configuration already in my Agent environment. Reuse my securely configured CAPYBROWSER_BASE_URL and CAPYBROWSER_API_KEY; never ask me to paste a key into chat. Check /version, authentication, and current pricing without creating a browser run. Wait for my actual web task. Use either MCP or HTTP to manage one run, and a CDP client for page actions. If I must sign in, give me that run's human-action link; after I return control, reconnect and inspect the page. Retrieve needed files before explicitly ending any run you created, then verify release, saved state, and settlement separately. A disconnected client does not end a run.