- Use cases
- AI agents
A real browser for Claude and Cursor.
Add one block of config and your assistant can browse, fill in forms and pull data from real web pages, right from the chat. The Browsa MCP server hands each request to a browser agent running real Chromium in the cloud and returns the answer with a link to watch it live.
100 free credits. No card, no subscription.
How it works.
Get an API key
Sign up and copy your
agt_live_key. You get 100 free credits once you verify your email.Add the server
Paste the config into Claude Desktop or Cursor and restart the app. It runs with
npx, so there is nothing else to install.Ask in plain language
Try "Go to news.ycombinator.com and tell me the top story." The assistant calls
run_taskand shows you the result.
{
"mcpServers": {
"browsa": {
"command": "npx",
"args": ["-y", "@browsa/mcp"],
"env": {
"BROWSA_API_KEY": "agt_live_…"
}
}
}
}
{
"mcpServers": {
"browsa": {
"command": "npx",
"args": ["-y", "@browsa/mcp"],
"env": {
"BROWSA_API_KEY": "agt_live_…",
"BROWSA_LLM_API_KEY": "sk-ant-…"
}
}
}
}
# The same kind of task the MCP server runs, from your own code from browsa import Client client = Client(api_key="agt_live_…") job = client.create_task( task="Go to news.ycombinator.com and return the top story", llm="browsa", ) job = client.wait_for_job(job.id) print(job.final_result, job.live_url)
Claude Desktop: claude_desktop_config.json. Cursor: ~/.cursor/mcp.json. Package: @browsa/mcp.
What you get.
Works in your MCP client
Claude Desktop, Cursor, Cline, Continue, Zed and other clients that speak MCP.
No model key needed
With only a Browsa key, tasks run on Browsa's model. Add your own Anthropic key to use it instead.
Watch it work
Every task returns a live view link, so you can see the browser while your assistant waits for the answer.
More than one-off tasks
The server also gives your assistant tools for schedules, output files, webhooks and saved identities.
Pricing.
Each request your assistant makes is an agent task. The MCP server itself is free and MIT licensed.
An agent task is a 5-credit browser, plus 1 credit, plus 1 credit for every 5 steps. Running it on Browsa's model adds 5 credits.
Free
Starter
Growth
Pro
Questions.
Which apps does it work with?
Claude Desktop, Cursor, Cline, Continue, Zed and other MCP clients. For Claude Desktop, the config goes in ~/Library/Application Support/Claude/claude_desktop_config.json on macOS or %APPDATA%\Claude\claude_desktop_config.json on Windows. For Cursor, use ~/.cursor/mcp.json.
Do I need an Anthropic or OpenAI key?
No. With only BROWSA_API_KEY set, tasks run on Browsa's model for 5 extra credits each. Add BROWSA_LLM_API_KEY to use your own key instead.
What tools does my assistant get?
run_task to do something in a browser; get_job, cancel_job and respond_to_job to follow up; create_schedule and list_schedules for recurring runs; list_job_files for output files; plus credits, webhook tools and list_identities.
What does it need on my computer?
Node.js 18 or later, for npx. The browser itself runs in Browsa's cloud, not on your machine.
Give your assistant a real browser.
100 free credits. No card, no subscription.