Give your agent a browser in one config block.

Paste the config for your client and Claude Code, Cursor, Claude Desktop or Windsurf can run browser tasks on stealth cloud browsers and read the result back.

Get an API key MCP docs

tools on the hosted server: 7
browser time inside a run, plus model tokens: $0.02 / hour
of credit for eligible new Google, GitHub or Microsoft signups: $1

Cloud, or your own Chrome.

The hosted server hands a whole task to an agent on a cloud browser and returns the result. The local stdio server, browser-harness-mcp, exposes click, type, read and screenshot on the Chrome you are logged into. For agents that would rather write their own browser code, see the CLI and skill.

claude mcp add -t http \  -H "Authorization: Bearer YOUR_API_KEY" \  browser-use https://api.browser-use.com/v3/mcp

terminal·Streamable HTTP · API key header

Which browser MCP server?

Pick by where the browser runs and who drives it.

Server Runs on How it works Best for Cost

Browser Use hosted MCP Browser Use cloudWhole task, result backMany tasks in parallel, off your machine$0.02 / browser-hour + model tokens

browser-harness-mcp Your machine, your logged-in ChromeStep by step, 23 toolsSites you are logged intoFree, MIT

Playwright MCP Your machine. Chrome, Firefox, WebKit, Edge. Your logged-in Chrome via its extensionStep by step, accessibility-tree snapshotsTests, with an isolated profile via --isolatedFree, Apache-2.0

Chrome DevTools MCP Your machine, Chrome only. Can attach to a running ChromeStep by step, plus traces, network, consoleDebugging performance, network or console errorsFree, Apache-2.0

Browserbase MCP Browserbase cloud. Self-host repo archived Jul 2026Step by step, natural-language act, observe, extractTeams already on BrowserbaseBrowserbase plan: 1 hr free, then from $20 / month

Matched eval, same agent, browser and judge on 100 tasks, mean score 0-100: Browser Harness (as a CLI) 60.4, Playwright MCP 60.0, Chrome DevTools MCP 61.0. A statistical tie.

Source: Browser Use MCP docs, browser-harness-mcp, Browser Harness README, Playwright MCP README, Chrome DevTools MCP README, Browserbase MCP setup, Browserbase MCP repo, Browserbase pricing, checked Oct 2026.

The seven tools

run_session: Create a session and run a task. Options: keep_alive, model, output_schema, profile_id.
get_session: Status, step count, cost so far, and the live URL.
send_task: A follow-up task to an idle keep-alive session.
stop_session: Stop the task, or destroy the whole session.
get_session_messages: The agent's actions, reasoning and results.
list_sessions: Recent sessions with status and cost.
list_browser_profiles: Browser profiles for logged-in tasks.

Questions people ask

What is the best MCP server for browser automation?

It depends on the job. Pick Playwright MCP for tests (isolated profile via --isolated), Chrome DevTools MCP to debug performance, network or console errors in your own app, and browser-harness-mcp to drive your own logged-in Chrome. For cloud browsers, pick the Browser Use hosted server to hand off whole tasks, or Browserbase MCP if you already run on Browserbase.

Which clients work?

Any MCP client that speaks streamable HTTP with headers. Claude Code, Claude Desktop, Cursor and Windsurf are documented. The local stdio server works with Claude Code, Devin, Cursor and other stdio clients.

Where do I get the key?

Create one at cloud.browser-use.com under Settings, API keys. Eligible new Google, GitHub or Microsoft signups get $1 of credit once, no card.

What does a task cost?

Model tokens plus browser time at $0.02 per browser-hour. On our hardest internal set the agent solved 82% of 106 tasks at 17¢ per solved task.

Can it return JSON?

Yes. Pass output_schema to run_session and the result matches your schema.