adk-browser

Browser automation tools for Rust Agent Development Kit (ADK-Rust) agents using WebDriver.

Overview

adk-browser provides 46 comprehensive browser automation tools that enable AI agents to interact with web pages, extract information, fill forms, take screenshots, and more. Built on the WebDriver protocol (Selenium), it works with any WebDriver-compatible browser.

Features

• 46 Browser Tools: Complete web automation toolkit for AI agents
• WebDriver Compatible: Works with Selenium, ChromeDriver, GeckoDriver, etc.
• ADK Integration: Tools implement adk_core::Tool for seamless agent integration
• Configurable: Filter tools by category based on agent needs
• Async: Built on Tokio for efficient async operations

Quick Start

Add to your Cargo.toml:

[dependencies]
adk-browser = "0.1"
adk-agent = "0.1"
adk-model = "0.1"

Prerequisites

Start a WebDriver server (e.g., Selenium):

docker run -d -p 4444:4444 -p 7900:7900 --shm-size=2g selenium/standalone-chrome:latest

Basic Usage

use adk_browser::{BrowserSession, BrowserToolset, BrowserConfig};
use adk_agent::LlmAgentBuilder;
use adk_model::GeminiModel;
use std::sync::Arc;

#[tokio::main]
async fn main() -> Result<(), Box<dyn std::error::Error>> {
    // Create browser session
    let config = BrowserConfig::new("http://localhost:4444");
    let session = BrowserSession::new(config).await?;

    // Create toolset with all 46 tools
    let toolset = BrowserToolset::new(session);
    let tools = toolset.all_tools();

    // Create AI agent with browser tools
    let model = Arc::new(GeminiModel::from_env("gemini-2.0-flash")?);

    let mut builder = LlmAgentBuilder::new("web_agent")
        .model(model)
        .instruction("You are a web automation assistant. Use browser tools to help users.");

    for tool in tools {
        builder = builder.tool(tool);
    }

    let agent = builder.build()?;

    Ok(())
}

Filtered Tools

Select only the tools your agent needs:

let toolset = BrowserToolset::new(session)
    .with_navigation(true)   // Navigate, back, forward, refresh
    .with_extraction(true)   // Extract text, links, HTML
    .with_interaction(true)  // Click, type, select
    .with_forms(false)       // Disable form tools
    .with_screenshots(true)  // Screenshots
    .with_javascript(false)  // Disable JS execution
    .with_cookies(false)     // Disable cookie tools
    .with_frames(false)      // Disable frame tools
    .with_windows(false)     // Disable window tools
    .with_actions(false);    // Disable advanced actions

let tools = toolset.selected_tools();

Available Tools (46 Total)

Navigation (6 tools)

Extraction (6 tools)

Interaction (6 tools)

Forms (5 tools)

Screenshots (3 tools)

JavaScript (3 tools)

Wait (4 tools)

Cookies (4 tools)

Frames (3 tools)

Windows (4 tools)

Actions (2 tools)

Element Selectors

Tools that target elements accept CSS selectors:

// By ID
"#login-button"

// By class
".submit-btn"

// By tag
"input[type='email']"

// By attribute
"[data-testid='search']"

// Complex selectors
"form.login input[name='password']"

Example: Web Research Agent

use adk_browser::{BrowserSession, BrowserToolset, BrowserConfig};
use adk_agent::LlmAgentBuilder;

let session = BrowserSession::new(BrowserConfig::new("http://localhost:4444")).await?;

let toolset = BrowserToolset::new(session)
    .with_navigation(true)
    .with_extraction(true)
    .with_screenshots(true);

let agent = LlmAgentBuilder::new("researcher")
    .model(model)
    .instruction(r#"
        You are a web research assistant. When asked about a topic:
        1. Navigate to relevant websites
        2. Extract key information using browser_extract_text
        3. Take screenshots of important content
        4. Summarize your findings
    "#)
    .tools(toolset.selected_tools())
    .build()?;

Configuration

let config = BrowserConfig::new("http://localhost:4444")
    .with_headless(true)          // Run headless (if supported)
    .with_timeout(Duration::from_secs(30))
    .with_implicit_wait(Duration::from_secs(10));

let session = BrowserSession::new(config).await?;

WebDriver Options

Works with any WebDriver-compatible server:

Examples

Run the included examples:

# Basic browser session
cargo run --example browser_basic

# AI agent with browser tools
cargo run --example browser_agent

# Full interactive example with all 46 tools
cargo run --example browser_interactive

# OpenAI-powered browser agent
cargo run --example browser_openai --features openai

License

Apache-2.0