> ## Documentation Index
> Fetch the complete documentation index at: https://docs.pre.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# Browser Agents

> Browser automation for humans and AI. Navigate, interact, and extract data from any website.

**Browser Agents for humans + AI.** Give the agent a URL and an instruction — it drives a real browser, follows the flow, and returns structured JSON validated against your schema. No Playwright plumbing, no selector maintenance.

<CardGroup cols={2}>
  <Card title="Quickstart" icon="play" href="/browser-agents/quickstart">
    Copy-paste examples in curl, TypeScript, Python, and MCP. Your first task in under a minute.
  </Card>

  <Card title="Get an API key" icon="key" href="https://pre.dev/projects/browser-agents">
    Grab a key from the pre.dev playground and set it as `PREDEV_API_KEY`.
  </Card>
</CardGroup>

## What Browser Agents can do

<CardGroup cols={2}>
  <Card title="Pull data off any site" icon="database">
    * Scrape pricing, listings, and contacts into JSON
    * Enrich CSVs with fresh web data
    * Read authed dashboards (Linear, Notion, Stripe)
  </Card>

  <Card title="Automate anything clickable" icon="hand-pointer">
    * Log in, navigate, fill forms
    * Submit bulk forms from a CSV
    * Step through multi-page flows end-to-end
  </Card>

  <Card title="Catch regressions before users do" icon="shield-check">
    * Run critical flows on a schedule
    * Verify checkout across plans
    * Flag pricing, UI, or copy drift
  </Card>

  <Card title="Give your agent a real browser" icon="plug">
    * Drop-in MCP for Claude Code & Cursor
    * URL + instruction → typed JSON
    * Replaces flaky Playwright code
  </Card>
</CardGroup>

## Benchmark

pre.dev Browser Agents passes **100 / 100** tasks in our public suite vs Browser Use Cloud's **97 / 100** — and it's **2.9× cheaper AND 2.8× faster**.

| Provider                   |     Pass rate |    \$ / task |   Total \$ | Avg time / task |
| -------------------------- | ------------: | -----------: | ---------: | --------------: |
| **pre.dev Browser Agents** | **100 / 100** | **\$0.0129** | **\$1.29** |      **12.7 s** |
| Browser Use Cloud          |      97 / 100 |     \$0.0372 |     \$3.72 |          35.7 s |

Same 100 tasks. Same JSON output schemas. Same uniform pass predicate.

* **[Interactive report](https://pre.dev/benchmark.html)** — radar chart, leaderboard, per-task drilldown with traces and screenshots.
* **[Raw data + reproduction](https://github.com/predotdev/browser-agents-benchmark)** — clone, set two API keys, `bun run bench`.

## Two ways to call it

<CardGroup cols={2}>
  <Card title="REST API" icon="code" href="/browser-agents/api/run-task">
    `POST https://api.pre.dev/browser-agent` with a tasks array. Sync, async, or live SSE streams. Up to 1,000 tasks per request.
  </Card>

  <Card title="MCP Tool" icon="plug" href="/browser-agents/mcp-tool">
    One-click install into Cursor, Claude Code, VS Code, or Windsurf. Your agent gets `browser_agent` as a tool — no orchestration code.
  </Card>
</CardGroup>

## Next

* [Quickstart](/browser-agents/quickstart) — runnable curl / TS / Python / MCP
* [REST API reference](/browser-agents/api/run-task)
* [MCP tool setup](/browser-agents/mcp-tool)