Docs · Getting started
From sign-in to your first task in five minutes
Human Browser runs a real Chromium in the cloud and does a task you describe in plain English, while you watch. One token works for the web app, MCP and the API.
Sign in and this page fills your token into every snippet. Until then they show hb_live_YOUR_TOKEN.
Get your token
Sign in on your account page with Google or email. Your API token (
hb_live_…) is created straight away and shown there. Keep it out of URLs, screenshots and public repos.Run a first task
The account page plays a short free demo, then lets you type a goal. Each of these three ran end to end on production; pick one to start with:
- Scrape a pageOpen https://books.toscrape.com and list the first 5 book titles on the page with their prices.Run it →
- Sign in to a siteGo to https://the-internet.herokuapp.com/login, sign in with username tomsmith and password SuperSecretPassword!, then tell me the exact message shown after login.Run it →
- Download a fileOpen https://the-internet.herokuapp.com/download, download the first .txt file in the list, and tell me its file name and size.Run it →
Connect Claude Code or Cursor (MCP)
Claude Code, one line in your terminal:
claude mcp add --transport http humanbrowser https://agent.humanbrowser.cloud/mcp \ --header "Authorization: Bearer hb_live_YOUR_TOKEN"Cursor: add this to
~/.cursor/mcp.json(or Settings → MCP → Add server):{ "mcpServers": { "humanbrowser": { "url": "https://agent.humanbrowser.cloud/mcp", "headers": { "Authorization": "Bearer hb_live_YOUR_TOKEN" } } } }Then ask your assistant something like “use humanbrowser to list the first 5 books on books.toscrape.com”. The token travels in the
Authorizationheader, never in the URL. Claude Desktop, Cline and other clients: MCP guide.Call it from code (A2A / REST)
Send the goal with
message/send. The reply arrives in a few seconds with a task id and a live viewer link; the run continues in the background.curl -s https://agent.humanbrowser.cloud/a2a \ -H "Authorization: Bearer hb_live_YOUR_TOKEN" \ -H 'Content-Type: application/json' \ -d '{"jsonrpc":"2.0","id":1,"method":"message/send","params":{"message":{"role":"user","messageId":"m1","parts":[{"kind":"text","text":"Open https://books.toscrape.com and list the first 5 book titles on the page with their prices."}]}}}'Poll
tasks/getevery few seconds until the state iscompleted; the answer is inmetadata.outcome.result.curl -s https://agent.humanbrowser.cloud/a2a \ -H "Authorization: Bearer hb_live_YOUR_TOKEN" \ -H 'Content-Type: application/json' \ -d '{"jsonrpc":"2.0","id":2,"method":"tasks/get","params":{"id":"t_…"}}'Logins without exposing passwords to the model, streaming, callbacks and error codes: API reference · A2A protocol details · OpenAPI · llms-full.txt for your agent.
Watch live and take control
Every task returns a viewer link (
viewerUrlin the API,viewer_urlover MCP). Open it to watch the browser as it works. Press Take control to use the mouse and keyboard yourself, for example to enter a code, then hand back. If the agent needs something only you know (a one-time code, a decision) it pauses and asks, in the viewer or through the API. Time under your control is not billed, and finished runs can be replayed.The viewer link includes a key that opens that session. Share it only with people who should see it.
Pricing
Pay as you go: $0.02 per agent step with the AI included, plus $0.10 per browser-hour. A typical task costs about $0.30, so a $9 pack covers roughly 30 tasks. Full pricing →