browser-agent

cyrup-ai's avatarfrom cyrup-ai

Autonomous browser agent for complex multi-step web tasks. WHEN: User needs complex web automation, multi-page workflows, form filling sequences, tasks requiring multiple browser actions. WHEN NOT: Simple single-action tasks like one click (use browser_click), one navigation (use browser_navigate), one extraction (use browser_extract_text).

⭐0stars🔀0forks📁View on GitHub🕐Updated Dec 26, 2025
[Web Automation]

When & Why to Use This Skill

The browser-agent is a sophisticated autonomous web automation tool designed to handle complex, multi-step browser tasks that exceed the capabilities of simple single-action tools. By leveraging AI to plan and execute workflows, it can navigate dynamic websites, manage sessions in the background, and interact with Single Page Applications (SPAs) to streamline repetitive and intricate web-based processes.

Use Cases

  • Multi-Step Web Research: Automatically navigating through various pages to find, aggregate, and summarize specific information, such as tracking down the latest software release notes across different repositories.
  • Complex Form Filling: Executing sequential data entry across multi-page web forms, including contact submissions and registration workflows that require logical progression.
  • Automated Data Extraction: Navigating deep into web portals to locate and extract specific datasets or documents that are not accessible via a single URL.
  • Background Task Execution: Running long-duration web workflows in the background with progress monitoring, allowing for parallel task management without manual intervention.
  • Dynamic Content Interaction: Interacting with modern, JavaScript-heavy websites and SPAs that require human-like navigation, clicking, and waiting for elements to load.
namebrowser-agent
description>
WHENUser needs complex web automation, multi-page workflows, form filling sequences, tasks requiring multiple browser actions.
WHEN NOTSimple single-action tasks like one click (use browser_click), one navigation (use browser_navigate), one extraction (use browser_extract_text).
version0.1.0

browser_agent - Autonomous Web Automation

Core Concept

mcp__plugin_kg_kodegen__browser_agent is an autonomous agent that performs complex multi-step web tasks. It:

  • Plans and executes browser actions autonomously
  • Handles dynamic content and SPAs
  • Supports session management for parallel tasks
  • Runs in background with progress monitoring

Uses session-based execution. Actions: PROMPT, READ, KILL.

Key Parameters

action (required)

  • PROMPT: Start agent with a task
  • READ: Check agent progress
  • KILL: Terminate agent session

task (required for PROMPT)

Natural language description of what to accomplish.

agent (optional, default: 0)

Agent slot number for parallel execution.

start_url (optional)

Initial URL to navigate to before starting task.

max_steps (optional, default: 10)

Maximum iterations the agent will perform.

await_completion_ms (optional, default: 600000)

Timeout in ms. 0 = fire-and-forget.

temperature (optional, default: 0.7)

LLM creativity (0.0-2.0).

max_actions_per_step (optional, default: 3)

Actions per iteration.

additional_info (optional)

Extra context for the agent.

Usage Examples

Start Agent Task

browser_agent({
  "action": "PROMPT",
  "task": "Find the latest Rust release notes and extract key features",
  "start_url": "https://www.rust-lang.org",
  "max_steps": 8
})

Form Filling Workflow

browser_agent({
  "action": "PROMPT",
  "task": "Fill out the contact form with name 'John Doe' and email 'john@example.com', then submit",
  "start_url": "https://example.com/contact",
  "temperature": 0.5,
  "max_actions_per_step": 2
})

Check Progress

browser_agent({
  "action": "READ",
  "agent": 0
})

Fire and Forget

browser_agent({
  "action": "PROMPT",
  "task": "Download the PDF from the resources page",
  "start_url": "https://example.com",
  "await_completion_ms": 0
})
// Returns immediately, agent works in background

Kill Agent

browser_agent({
  "action": "KILL",
  "agent": 0
})

Response Format

{
  "agent": 0,
  "task": "Find latest Rust release notes",
  "steps_taken": 5,
  "completed": true,
  "summary": "Found release notes for Rust 1.75...",
  "history": [
    {"step": 1, "actions": ["navigate"], "summary": "Navigated to rust-lang.org"},
    {"step": 2, "actions": ["click"], "summary": "Clicked releases link"}
  ]
}

When to Use What

Complexity Tool Use Case
Single action browser_click/type/etc Click a button, type text
2-3 actions Manual tool calls Login form, simple search
4+ actions browser_agent Multi-page workflows
Research browser_research Topic research with summaries

Configuration Guide

Scenario Settings
Web research temperature: 0.8, max_steps: 12
Form filling temperature: 0.5, max_actions_per_step: 2
Complex workflow max_steps: 15-20
Quick task max_steps: 5, await_completion_ms: 60000

Remember

  • Use for complex multi-step tasks only
  • Simple tasks are faster with direct tool calls
  • Agent runs in background (check with READ)
  • Provide clear task descriptions
  • Include start_url when known
  • Kill agent when done to free resources