minion-agent

MCP.Pizza Chef: femto

This is a Python toolkit for people who write code, not an app you install and click. In a few dozen lines you assemble an assistant that drives a browser, writes and runs programs, plans multi-step work, and picks up tools from any MCP server you point it at. You bring your own model access, with Azure OpenAI documented as the recommended route. The project has been quiet since January 2026.

Coding

Web/Research

Use This MCP client To

Build an assistant that shops and compares prices for me Run a deep research task across many web pages Have an agent write and test a small program Add tools from an MCP server to my own agent Automate a browsing task I repeat every week

README

Minion Agent

A powerful agent framework with enhanced capabilities including browser automation, code execution, MCP tool support, and deep research. Now defaults to EXTERNAL_MINION_AGENT framework for superior performance and functionality.

🎬 Demo Videos

Quick Start

Minion Agent now defaults to the powerful EXTERNAL_MINION_AGENT framework, providing enhanced code execution, browser automation, and advanced planning capabilities out of the box.

Installation

pip install minion-agent-x

Or from source

minion-agent FAQ

Is this project still moving?

Slowly. The last code change was January 2026, about seven months before this listing, and no replacement project is named.

Is this an app I can install and click?

No. It is a Python library you write code against, so you need to be comfortable in a code editor.

Do I need a paid key?

Yes. You supply your own model access, and Azure OpenAI is the option the project documents most fully.

Which apps does it work in?

None directly. Instead of plugging into Claude Desktop or Cursor, this is the thing that consumes MCP servers itself.

Can I use this to research a topic across dozens of pages?

Yes. Deep research is one of its built-in modes, and the authors demonstrate it in a video.

Can it control a real browser?

Yes. Browser automation is built in, so your agent can click through sites rather than only reading text.

How hard is the setup?

Developer level. You install it with pip, write a short Python file, and put your model keys in an environment file.