epusapati/mcp

RAG Architecture: (Store, Retrieve, Generate)  Documents will be broken into chunks and stored as searchable vectors in a database.  When you ask a question, the most relevant chunks are pulled out and handed to the AI.  The AI reads those chunks and answers your question using that contex. MCP Architecture: ( Connect, Discover,Act) • Your tools (APIs, databases, apps) each expose an MCP server — a standard interface the AI can talk to. • The AI asks “what can you do?” and the server lists its available actions in real time. • The AI calls the right tool, gets a live result, and completes the task

MCP Components:

1. Host Application The outer shell that runs the AI — for example Claude Desktop or VS Code or Cursor. It starts everything, manages the session, and owns the user interface.
2. MCP Client Lives inside the host. It is the communication bridge that sends tool requests to servers and passes results back to the AI. One client per server connection.
3. Protocol Layer The universal language — all messages between client and server use the same JSON format, sent over either stdio (local) or HTTP (remote). This is what makes any client work with any server.
4. MCP Server The component you install or build that wraps an external service — a database, API, or filesystem — and exposes it to the AI in standard MCP format. It handles authentication and executes requests. (e.g., Google Drive, GitHub, Slack)
5. Tools Actions the AI can call — for example send_email or run_query. Each tool has a name, description, and parameters. The AI reads the description and decides when to use it.
6. Resources Data the AI can read — files, database records, documents. Unlike tools, resources are not called with parameters; they are accessed like documents to pull context into the conversation.
7. Prompts Reusable instruction templates stored on the server. Instead of typing a long instruction every time, a user selects a named prompt like code_review and the full instruction loads automatically.