Mini-Agent-TS

English | 简体中文

This project is a TypeScript implementation of MiniMax's open-source Mini-Agent project.

Mini-Agent-TS is a terminal LLM Agent that supports extending capabilities through Agent Skills and MCP (Model Context Protocol). It is compatible with both Anthropic and OpenAI protocols, featuring native file operations and command execution capabilities, making it an all-purpose AI assistant for developers in the terminal environment.

---

✨ Core Features

• 🔄 ReAct Mode: Supports ReAct Agent loop mechanism for multi-step reasoning and tool calling to complete complex tasks.
• 🧠 Interleaved Chain of Thought: Tightly integrates Agent's reasoning process with tool calls.
• 🔌 MCP Protocol Support: Easily connect to external tool ecosystems and extend Agent capabilities.
• 🛠️ Agent Skills: Customize Agent skills through professional knowledge bases, workflows, and toolsets to build domain experts.
• 🌐 Custom Providers: Supports Anthropic and OpenAI SDKs, with freedom to connect to any LLM provider compatible with these protocols.

---

📂 Project Structure

Mini-Agent-TS/
├── src/
│   ├── agent.ts           # Agent core logic
│   ├── cli.ts             # CLI entry point
│   ├── config.ts          # Configuration loading and parsing
│   ├── llm-client/        # LLM client adapters
│   ├── schema/            # Data model definitions
│   ├── skills/            # Skills loader
│   ├── tools/             # Built-in toolset
│   └── util/
├── config/
│   ├── config.yaml        # Main configuration file
│   └── mcp.json           # MCP server configuration
├── skills/                # User-defined skills directory
├── tests/                 # Test files
└── logs/                  # Runtime logs (generated when logging is enabled)

---

🛠️ Quick Start

1. Clone Project and Install Dependencies

Run the following commands to deploy the project locally:

# Clone repository
git clone [https://github.com/Code-MonkeyZhang/Mini-Agent-TS.git](https://github.com/Code-MonkeyZhang/Mini-Agent-TS.git)

# Enter project directory
cd Mini-Agent-TS

# Install dependencies
npm install

2. Build and Link to System Commands

npm run build && npm link

Project Configuration

Initialize the configuration file with your API information:

# Copy example configuration
cp config/config-example.yaml config/config.yaml

# config/config.yaml

# Enter your API Key
api_key: "YOUR_API_KEY_HERE" # Replace with your LLM provider API Key
api_base: "https://api.minimax.io/anthropic" # Replace with your base url

# Model and provider SDK format
model: "MiniMax-M2"
provider: "anthropic" # "anthropic" or "openai"

# Logging configuration (optional)
enableLogging: false # Set to true to enable logging, logs saved in logs/ folder at project root

Specify Workspace

If you want the Agent to operate only within a specific project directory (to prevent accidental deletion or modification of other files), use:

🔌 MCP Servers

This project supports adding external tools to the Agent via the MCP protocol. The following example shows how to add a time server: Edit config/mcp.json:

{
  "mcpServers": {
    "time-server": {
      "command": "uvx",
      "args": ["mcp-server-time"],
      "description": "Provides current time query tools"
    }
  }
}

🧠 Agent Skills

This project supports Agent Skills, allowing users to add "operation manuals" with specific functionality to the Agent. To add a skill, you need to create a skills directory in the project root or a specified location. Place your skill files in that directory. Also ensure the correct skills path is enabled in config.yaml:

tools:
  skillsDir: "./skills"

---

🤝 Contributing & Feedback

Issues and Pull Requests are welcome to improve this project.

Made with ❤️ by Code-MonkeyZhang

---

📚 Reference Documentation

The implementation of this project references the following official documentation:

• OpenAI API Reference
• Anthropic Messages API
• Model Context Protocol Docs
• Agent Skills Documentation