Report Abuse

Basic Information

Minion Agent is a lightweight agent framework intended for developers who want to build and run autonomous agents that can use a browser, integrate with Model Context Protocol (MCP) tools, perform automatic instrumentation, create and follow plans, and run deep research workflows. The project exposes a MinionAgent class and an AgentConfig configuration object to specify model ids, model types, tools, instructions, and agent-specific arguments. It integrates with the smolagents framework and supports multiple model backends referenced in examples, including Azure and OpenAI server models and a LiteLLM option. The README includes usage examples, demo videos, example scripts for browser use, managed agents, deep research and reasoning, and instructions to install via pip or from source. The repository is packaged for development and includes guidance on environment variables and a development workflow.

Links

App Details

Features
The repo documents a set of concrete features for building and running agents. Configurable AgentConfig parameters let you set model_id, name, description, instructions, tools, model_type, model_args, agent_type and agent_args. MCP tool support includes both local command-based MCP tools and SSE-based remote MCP tools, with examples showing how to supply commands or URLs for MCPTool. Planning support is provided via smolagents through an agent_args planning_interval that periodically creates and updates plans. The project supplies example scripts for browser interactions, managed agents, deep research, and reasoning. There are demo videos and a dedicated deep research documentation page. Development convenience includes pip installation, editable installs, a .env pattern for API keys, and a security warning about trusting MCP servers.
Use Cases
Minion Agent helps developers prototype and deploy autonomous agents that need multi-step planning, external tooling, or browser automation. It streamlines configuration of model backends and exposes a unified AgentConfig so teams can swap model types and provide model-specific arguments without changing agent logic. MCPTool support enables connecting agents to filesystem or remote MCP servers to extend capabilities with external code or data, and SSE-based tooling allows remote MCP servers where appropriate. Planning integration in smolagents automates periodic plan generation and re-evaluation of tasks. The included examples, demos, and development instructions make it easier to reproduce research workflows, build research assistants, test browser-enabled agents, and iterate locally with editable installs and environment variable configuration.

Please fill the required fields*