Use the current hosted Browser Use MCP endpoint with Claude Code by registering https://api.browser-use.com/v3/mcp and sending your Browser Use API key in the x-browser-use-api-key HTTP header. Then run /mcp in Claude Code to verify the connection. Hosted sessions execute in Browser Use’s cloud browser; the local alternative runs a browser on your machine and gives Claude step-by-step control.
What you are connecting
Model Context Protocol (MCP) lets Claude Code call tools exposed by another server. Browser Use offers two materially different arrangements:
![Claude AI for Beginners Bible: [5 in 1] The Ultimate Guide to Automate Your Work, Save Hours Every Week, and Use AI for Real-World Results](https://m.media-amazon.com/images/I/415+fSJacsL._SL160_.jpg)
- Hosted MCP: Claude sends a task to Browser Use Cloud. The agent operates a cloud browser and returns the session result.
- Local MCP: A stdio server runs on your computer. Claude controls a browser locally, while you supply the model credentials required by the local package.
The current hosted endpoint in Browser Use’s product page and MCP documentation is https://api.browser-use.com/v3/mcp. An older repository integration still shows the unversioned https://api.browser-use.com/mcp; use the v3 endpoint for a new setup and treat the older value as a likely source of failures in existing configurations.
Before connecting any server, follow Anthropic’s warning: “Verify you trust each server before connecting it. Servers that fetch external content can expose you to prompt injection risk.” A browser agent reads instructions from arbitrary pages, so do not give it secrets or permissions that are unnecessary for the task.
Prerequisites and safe preparation
- Install and authenticate Claude Code, then confirm the
claudecommand works in your terminal. - Create a Browser Use API key in your Browser Use account settings.
- Keep the key in a password manager or environment-specific secret store. Never commit it to a repository, paste it into a prompt, or publish it in shell history or documentation.
- Decide whether cloud execution is acceptable for the pages and data you will visit.
Browser Use’s official pages are the authority for current account steps and availability: MCP server documentation and Browser Use MCP product page.
Set up the hosted Browser Use MCP server
- Open Browser Use account settings and create an API key.
- In a terminal where Claude Code is installed, register the remote HTTP server. Replace the placeholder with your own key:
claude mcp add -t http -H "x-browser-use-api-key: YOUR_API_KEY" browser-use https://api.browser-use.com/v3/mcpThe command stores the server under the name browser-use. The key is transmitted as an HTTP header rather than as part of the URL.
- Start (or restart) Claude Code in the project where you want to use the server.
- Run
/mcp. Claude Code’s documentation and support FAQ use this command to display MCP connection status. The Browser Use server should appear as connected; if it does not, troubleshoot the registration before attempting tasks.
Once connected, ask Claude to inspect the available Browser Use tools rather than assuming a fixed list. The current product page lists run_session, get_session, send_task, stop_session, get_session_messages, and list_sessions. The MCP documentation also lists list_browser_profiles. Tool visibility can vary by account or client version.
Run a first cloud-browser task
Start with a non-sensitive page and a bounded instruction. For example:
The Tool Desk
Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Open https://example.com, report the page title and the first heading, then stop.Ask Claude to explain which MCP tool it is calling and to return the session identifier. For longer work, a useful pattern is:
Rank #2
- Create a session with a concise objective.
- Poll or retrieve the session with
get_session. - Read progress or results with
get_session_messages. - Send a corrective instruction with
send_taskif the page state changes. - Call
stop_sessionwhen the task is complete or no longer needed.
Use list_sessions to find active work after a terminal disconnect. Keep tasks explicit about the allowed domain, actions, stopping condition, and information Claude may return. Never instruct an agent to defeat a CAPTCHA, bypass access controls, or enter credentials unless your organization has approved that workflow.
Hosted versus local: which should you choose?
| Characteristic | Hosted MCP | Local MCP |
|---|---|---|
| Browser location | Browser Use cloud | Your computer or controlled local environment |
| Agent interaction | Task handoff to a cloud agent | Claude controls the browser step by step through stdio |
| Primary credential | Browser Use API key in HTTP header | LLM key such as OPENAI_API_KEY or ANTHROPIC_API_KEY, plus local package configuration |
| Operational trade-off | Minimal local browser setup; data and execution leave your machine | More control over local data and browser; you maintain the runtime |
| Cost note | Browser time is billed by Browser Use, in addition to model tokens | The repository describes the server as free/self-hosted, but an LLM key and its provider costs are still required |
For hosted usage, Browser Use currently advertises $0.02 per hour for browser time inside a run, plus model tokens (checked September 29, 2026). Prices can change, so verify the product page before budgeting. Browser Use also reports an 82% solve rate on 106 tasks at 17¢ per solved task on its hardest internal set. That is a vendor-reported internal benchmark, not independent validation or a guarantee for your workload.
Set up the local stdio server
The official Browser Use integrations repository gives this launch command:
uvx --from 'browser-use[cli]' browser-use --mcpRun it in the environment where Claude Code can start stdio MCP processes, then follow Claude Code’s current MCP configuration guide for registering a local command at project or user scope. The repository states that local automation needs an LLM key, including OPENAI_API_KEY or ANTHROPIC_API_KEY. Package flags and dependencies may change, so check the current integration instructions before pinning this command in automation.
Choose local when the browser must reach resources available only on your network, or when you need direct control of the browser profile. Choose hosted when you want Browser Use to manage the browser runtime and can send the task and page data to its cloud service.
Configuration, scope, and secret handling
Claude Code supports MCP configuration at project and user scope. Put shared, non-secret server configuration in the appropriate configuration file, but inject API keys through the secret mechanism recommended by the current Claude Code MCP guide. Restrict a project-scoped server to repositories that actually need it. Rotate a key immediately if it appears in a commit, terminal transcript, issue, or chat export.
Browser pages can contain hostile text that looks like instructions. Treat page content as untrusted input: ask Claude to summarize or extract data, not to obey instructions found on the page. Require confirmation before sending messages, making purchases, changing account settings, downloading files, or submitting forms.
Quick wins for a faster PC:
Clear out junk files and repair common Windows errorsFree Scan →Scan for outdated or missing drivers - takes under a minuteDriver Scan →Repair Windows errors before they cause bigger problemsFix Now →Troubleshooting common failures
“Server not found” or no entry in /mcp
Check that the command used -t http, the exact v3 endpoint, and the server name browser-use. Restart Claude Code after changing registration. Copy the command from a plain terminal to avoid smart quotes.
Authentication or unauthorized errors
Generate a fresh Browser Use key, ensure there are no extra spaces, and confirm the header is exactly x-browser-use-api-key. Do not put the key in the endpoint URL. If the key was exposed, revoke it and create another.
The connection works but tools differ
Tool availability is not guaranteed to be identical across clients or accounts. Ask Claude to enumerate the server’s tools, and use the names returned by the live connection. Do not copy an older tutorial that assumes the legacy endpoint or tool names.
Local server exits immediately
Verify that uvx is installed, the Browser Use package extras resolve, and the required LLM environment variable is present. Run the command manually to read the startup error, then apply the repository’s current installation guidance.
Tasks stall or consume unexpected budget
Use a narrow goal, a clear stop condition, and short follow-up instructions. Stop abandoned sessions. For hosted runs, remember that browser time and model tokens are separate cost components.
Unexpected actions on a webpage
Stop the session, inspect messages, and tighten the prompt’s allowed actions. Assume external page text may be a prompt-injection attempt. Use a separate browser profile and least-privilege credentials for testing.
Performance and reliability practices
- Break a large workflow into checkpoints so a failed step can be retried without repeating everything.
- Record session IDs and final outputs in your own logs, excluding secrets and sensitive page content.
- Prefer deterministic selectors, explicit URLs, and verifiable outcomes over “figure it out” instructions.
- Use local execution for private intranet pages; use hosted execution for disposable public-page research.
- Review Browser Use pricing and tool documentation before production rollout because endpoints, tools, and rates are vendor-maintained and can change.
Or skip the browser setup
If you only need a clean image or PDF of a webpage—not an agent that clicks through a workflow—ScreenshotNeo is a simpler alternative. It removes cookie and consent banners, newsletter popups, and chat widgets before capture; bot checks, blank pages, failed loads, timeouts, and cache hits are not billed, with the result identified by X-Page-Verdict and X-Billed headers. Its MCP server gives Claude, Cursor, and other MCP clients take_screenshot, get_page_info, and capture_pdf tools.
Using the API is one GET request. See the ScreenshotNeo documentation for all options:
curl -G "https://api.screenshotneo.com/v1/shot" -d access_key=YOUR_API_KEY --data-urlencode url=https://stripe.com -o shot.webpPython:
import requests
r = requests.get("https://api.screenshotneo.com/v1/shot", params={"access_key": "YOUR_API_KEY", "url": "https://stripe.com"}, timeout=90)
open("shot.webp", "wb").write(r.content)Node.js:
const q = new URLSearchParams({ access_key: 'YOUR_API_KEY', url: 'https://stripe.com' });
const res = await fetch(`https://api.screenshotneo.com/v1/shot?${q}`);ScreenshotNeo includes full-page and element captures, lazy-image loading, device presets and custom viewports, dark mode, retina scale, PDF controls, custom CSS and JavaScript, clicks, waits, blocked resources, headers, cookies, user agents, authorization, timezone, geolocation, transparent backgrounds, resizing, TTL caching, signed links, asynchronous webhooks, bulk capture of up to 100 URLs per call, usage reporting, and an OpenAPI specification. Every feature is on every plan. The Free plan includes 1,000 screenshots per month with no card; paid plans start at $5 for 3,000. Create a free ScreenshotNeo account.
FAQ
Can I use both hosted and local servers?
Yes. Register them with distinct names and choose the one appropriate to each project and data boundary.
Does MCP make a browser workflow deterministic?
No. Page changes, authentication, timing, and model decisions can still affect results; use checkpoints and verification.
Is the old unversioned endpoint always broken?
Not necessarily, but current official product and documentation pages specify the v3 endpoint. Treat the old URL as legacy until Browser Use confirms otherwise.
Recommended Free Tools
Frequently Asked Questions
Can I use both hosted and local servers?
Yes. Register them with distinct names and choose the one appropriate to each project and data boundary.
Does MCP make a browser workflow deterministic?
No. Page changes, authentication, timing, and model decisions can still affect results; use checkpoints and verification.
Is the old unversioned endpoint always broken?
Not necessarily, but current official product and documentation pages specify the v3 endpoint. Treat the old URL as legacy until Browser Use confirms otherwise.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
Free tools Windows power users keep installed
One-click scans. No signup required.
