Browser Use (Cloud)

Overview

The Browser Use tool runs a browser automation task through the hosted Browser Use cloud API. Describe what the agent should do in plain language, and it navigates the web, fills forms, and extracts data for you. For the self-hosted variant that runs inside an E2B sandbox (with session reuse and screenshots), see Browser Use (Sandbox).

Key Features

BROWSER_USE_RUN_TASK
- Create a Browser Use task from a natural-language task description and, by default, wait for it to finish
- max_steps caps the number of agent steps (default: 100)
- wait_for_completion (default: true) polls the task until it finishes; set it to false to return right after the task is created
- timeout_seconds (default: 300) bounds how long the step waits for completion

Authentication

Set the api_key config to a Browser Use API key (secret type BROWSER_USE). Leave it blank to run on Jinba Credit instead: the step then reserves credits according to max_steps (320 credits at the default of 100). Note: Treat API keys as sensitive information and never commit them to public repositories.

Usage Examples

Example: Extract Data from a Website

- id: browse
  name: browse
  tool: BROWSER_USE_RUN_TASK
  config:
    - name: api_key
      value: "{{secrets.BROWSER_USE}}"
  input:
    - name: task
      value: Open https://news.ycombinator.com and return the titles and URLs of the top 5 stories as a JSON array.
    - name: max_steps
      value: 30
    - name: timeout_seconds
      value: 300

- id: summarize
  name: summarize
  tool: OPENAI_INVOKE
  config:
    - name: version
      value: gpt-4
  input:
    - name: prompt
      value: |
        Summarize these stories in one paragraph:
        {{steps.browse.result.output}}

Example: Start a Task Without Waiting

- id: start_task
  name: start_task
  tool: BROWSER_USE_RUN_TASK
  input:
    - name: task
      value: Visit https://example.com and report the page title.
    - name: wait_for_completion
      value: false

Notes

result contains task_id, status, output (the agent’s final answer as text), and steps — one entry per agent step with number, memory, evaluationPreviousGoal, nextGoal, url, screenshotUrl, and actions. With wait_for_completion: false, only task_id and status: created are returned.
Completion is polled every 3 seconds. A finished task reports status: finished or status: failed; if timeout_seconds elapses first, the step returns status: timeout together with an error, so the task keeps running on the Browser Use side.
The task text is masked for personal information according to your organization’s settings before it is sent to Browser Use.

​Overview

​Key Features

​Authentication

​Usage Examples

​Example: Extract Data from a Website

​Example: Start a Task Without Waiting

​Notes