Sentinel Core Agent

🚀 Sentinel Core Agent Overview

This document provides a summary of the client.py and server.py files in this repository.

🚀 Quick Start

💻 Usage Examples

Basic Usage

This README gives you a high - level understanding of how the client.py and server.py files work in the Sentinel Core Agent project.

✨ Features

Client.py

The client.py file manages the interaction between the user, the Large Language Model (LLM), and various tools. It initializes and coordinates chat sessions, where the LLM can use tools to answer user questions. The client handles server connections, tool execution (with a retry mechanism), and communication with the LLM provider.

Key Components:

• Configuration: Loads environment variables (including the LLM API key) and settings from the server configuration file.
• Server: Manages the connection to the MCP (Micro - service Communication Protocol) server, lists available tools, executes tools with a retry mechanism, and cleans up resources.
• Tool: Represents a tool and its attributes (name, description, input mode) and provides a method to format tool information for use by the LLM.
• LLMClient: Manages communication with the LLM, interacting using either the Azure OpenAI or Google Gemini models.
• ChatSession: Coordinates the interaction between the user, the LLM, and tools. It processes LLM responses, executes tools (if necessary), and manages the main chat session loop.

The client sets a system message that provides the LLM with available tools and instructions on using them. It then enters a loop where it gets user input, sends it to the LLM, and processes the LLM's response. If the response contains a tool call, the client executes the tool and returns the result to the LLM. This process continues until the user exits the chat session.

Server.py

The server.py file implements an MCP server with various tools, including file system operations, web scraping, and AI search capabilities. It uses the fastmcp library to create and run the server.

Key Tools:

• is_file_folder_present: Checks if a file or folder exists.
• current_datetime: Returns the current date and time.
• browser_ai_search: Searches the web using an AI agent (Brave Search) and returns the response.
• read_file: Reads the contents of a file.
• write_file: Writes content to a file (in both text and binary modes).
• web_page_scrapper: Scrapes web page content and returns it in Markdown format. It can also index the content for vector search.
• get_all_vector_indexes: Retrieves all vector embedding indexes in the current directory.
• search_via_index: Searches a query through a vector embedding index.

The server also initializes an asynchronous web crawler (crawl4ai) and sets up an embedding model for content vectorization. The server's lifecycle is managed using an asynchronous context manager to ensure the correct startup and shutdown of the crawler.

The server uses mcp.run(transport="stdio") to start the MCP server loop, allowing it to receive and process tool calls from clients.