Skip to content

README_EN.md

Latest commit

 

History

History
154 lines (113 loc) · 5.08 KB

README_EN.md

File metadata and controls

154 lines (113 loc) · 5.08 KB

🔧 Agent CLI - MCP Protocol Command Line AI Tool

  • A lightweight Rust-based MCP client
  • A command-line AI tool based on the MCP (Model Context Protocol) protocol, providing streaming chat interaction and tool calling functionality.
  • Supports NapCat connection to QQ

中文版本.

Examples

Features

  • ✨ Real-time streaming chat responses
  • ✨ Supports MCP tool calling and reasoning process display
  • ✨ Configurable MCP server connections
  • ✨ Built with Rust, high performance and reliable
  • ✨ Supports command-line interactive interface
  • ✨ Acts as a server to connect with NapCat and respond to QQ @ messages

🔧 Internal Tools

The Agent CLI includes the following built-in internal tools that can be used directly in chats:

Tool Name Description Main Functions
filesystem File system operations tool Read, write, list files and directories. By default, it can only operate on files within the current working directory.
get_best_tool Get best tool recommendation Analyze user requirements and recommend the most suitable available tools.
choose_tool Tool selector Inform the system and users about the most appropriate tool to use (typically called internally by get_best_tool).

📦 Installation Guide

Install from Source

  1. Ensure Rust is installed

  2. Clone the repository:

    git clone https://github.com/your-repo/agent-cli.git

  3. Build the project (NapCat is not compiled by default, need to add parameter --features napcat):

    cd agent-cli
cargo build --release --features napcat

  4. The binary file is located at target/release/agent-cli

  5. Rename config_temp.json to config.json, fill in your api_key registered on the deepseek official website, ensure you have both config.json and log4rs.yaml files locally If you need to use napcat, run with the parameter --napcat, then rename napcat_temp.toml to napcat.toml, it will only process messages sent by the target_qq configured

💬 Usage Instructions

Basic chat interaction:

agent-cli -p "Your question or instruction"

⚙️ Configuration Method

Configuration file is located at config.json, specific configuration reference config_temp.json file

Parameter Description

  • --prompt User input, if not provided, enters command-line interactive UI mode
  • --stream Whether to use streaming, defaults to true
  • --use_tool Whether to use tools, defaults to true
  • --wait Wait mode, defaults to false. When true, the program processes standard input in a loop, with no context preservation between conversations
  • --remote Start remote WebSocket server, specify listening address (e.g., 127.0.0.1:8080)

🌐 Remote Module - External Integration Guide

The Agent CLI provides a Remote module that allows external applications to interact with the AI model through WebSocket protocol. This module supports multiple input types and configuration options, making it easy to integrate into other systems.

Quick Start

  1. Start Remote Server:

    agent-cli --remote 127.0.0.1:8080

  2. Client Connection Example (Python):

    import asyncio
import websockets
import json

async def send_request(request_data):
    async with websockets.connect('ws://127.0.0.1:8080') as websocket:
        request_json = json.dumps(request_data)
        await websocket.send(request_json)
        response_data = await websocket.recv()
        return json.loads(response_data)

# Send request
response = asyncio.run(send_request({
    "request_id": "test_001",
    "input": {"Text": "Hello"},
    "stream": False,
    "use_tools": True
}))
print(response)

Detailed Protocol Documentation

Complete communication protocol documentation: remote_protocol.md

Documentation includes:

  • Complete protocol specifications
  • All message format definitions
  • Multiple input type support (text, images, files, instructions, etc.)
  • Configuration options explanation
  • Usage examples
  • Client implementation guides (Python, JavaScript, etc.)
  • Error handling and performance recommendations

Main Features

  • Multiple Input Types: Supports text, images (base64), files, structured instructions
  • Streaming Responses: Supports real-time streaming output
  • Tool Calling: Configurable MCP tool usage
  • Configuration Overrides: Supports request-level custom configuration
  • Token Statistics: Returns detailed token usage information

Integration Scenarios

  • Web Application Backend: As an AI service provider
  • Desktop Applications: Integrate AI functionality
  • Automation Scripts: Batch processing tasks
  • Monitoring Systems: Intelligent alert analysis
  • Educational Tools: Intelligent tutoring systems

👨‍💻 Development Guide

Build

cargo build

Run Tests

cargo test

or double click file “运行Target.bat”

Log Settings

Set log level and output in log4rs.yaml

📜 License

GPL_V3