Mobile Proxy for Browser Use: AI Agent Setup

An agent run is not one request. A single task like "find this product and compare prices" produces dozens of navigations, clicks, and form fills over several minutes, all from the same IP. If that IP is a datacenter address, anti-bot systems score the whole session as hosting-originated traffic and respond with CAPTCHAs, Cloudflare interstitials, or silent blocks. The agent then burns LLM steps trying to click through challenge pages it cannot pass.

The project is explicit that this is out of scope for the library. The Browser Use README's FAQ answers "How do I solve CAPTCHAs?" with: "For CAPTCHA handling, you need better browser fingerprinting and proxies", and points to its paid cloud for stealth browsers and proxy rotation. If you run the open-source library on your own infrastructure, the IP layer is yours to solve - and that is exactly what a dedicated 4G/5G mobile proxy does. Carrier IPs sit behind CGNAT, shared with thousands of real subscribers, so blocking one is expensive for the target site in false positives.

Browser Use exposes a typed ProxySettings model (defined in browser_use/browser/profile.py) with four fields: server (an http:// or socks5:// proxy URL), bypass (comma-separated hosts to skip), and username / password for authenticated proxies. Under the hood the server becomes Chromium's --proxy-server launch flag and credentials are supplied through the CDP auth challenge, so put them in the dedicated fields, not in the URL.

pip install browser-use   # or: uv add browser-use

# From your mobileproxies.org dashboard you need:
# host: proxy.mobileproxies.org   HTTP port: 8000
# username: u_4a9c                password: p_2X7q

import asyncio
from browser_use import Agent, Browser, ChatBrowserUse
from browser_use.browser import ProxySettings

browser = Browser(
    headless=True,
    proxy=ProxySettings(
        server="http://proxy.mobileproxies.org:8000",
        username="u_4a9c",
        password="p_2X7q",
        bypass="localhost,127.0.0.1",
    ),
)

agent = Agent(
    task="Open https://api.ipify.org and report the IP shown",
    browser=browser,
    llm=ChatBrowserUse(),
)

asyncio.run(agent.run())

Multi-step tasks need a stable identity. If your IP rotates mid-task - the default on many rotating residential pools - a login started on one IP finishes on another, and risk engines flag the session. A dedicated mobile proxy is sticky by design: the slot holds one carrier IP until you explicitly rotate it. Pair that with Browser Use's keep_alive option ("keep browser alive after agent run") and a persistent user_data_dir, and consecutive agent runs share the same IP, cookies, and browser profile - which is exactly what login-heavy flows like account creation and management require.

# One Browser per agent, each on its own dedicated slot
browser_a = Browser(
    proxy=ProxySettings(
        server="http://proxy.mobileproxies.org:8000",
        username="u_4a9c",
        password="p_2X7q",
    ),
    user_data_dir="~/.config/browseruse/profiles/agent-a",
    keep_alive=True,  # browser survives between agent runs
)

agent = Agent(task="Log in and open the orders page",
              browser=browser_a, llm=ChatBrowserUse())
await agent.run()

# Next task reuses the same carrier IP, cookies and profile
followup = Agent(task="Export this month's order list",
                 browser=browser_a, llm=ChatBrowserUse())
await followup.run()

Running agents in parallel? Give each its own Browser instance, its own user_data_dir, and its own proxy slot. Two agents behind one IP with identical fingerprints look like one bot pretending to be two people; two dedicated slots look like two unrelated phones.

Frequently asked questions

Sources

• • Browser Use Docs - Browser settings (proxy parameter, ProxySettings fields)
• • Browser Use Docs - All browser parameters (proxy, keep_alive, user_data_dir; Browser = BrowserSession)
• • Browser Use Docs - Browser basics (passing browser= to Agent)
• • GitHub - browser-use/browser-use (stars, v0.13.3 release, CAPTCHA FAQ)
• • GitHub - browser_use/browser/profile.py (ProxySettings model, keep_alive)
• • GitHub - tests/ci/browser/test_proxy.py (--proxy-server flag, CDP auth handling)