MCP-Server-LLM-Tool-Consumer

This project implements a robust bridge between a Deterministic Core (Python logic) and a Probabilistic Brain (Gemini LLM) using the Model Context Protocol (MCP) logic.

1. Technical Objective

The goal is to build a reliable microservice (the MCP Server) that performs logic without AI interference, while a smart but erratic LLM (the Client) identifies when to call those functions based on natural language input.

---

2. Architecture & Data Flow

The system follows a strict decoupling of responsibilities:

1. Discovery: The Client queries the Server for available tools and resources via stdio.
2. Planning: The Client passes the user's prompt and the tool's JSON schema to Gemini.
3. Execution: Gemini returns a tool request; the Client executes it against the Server's deterministic logic.
4. Final Response: The Client returns the raw tool result to the LLM for a natural language summary.

---

3. Tool & Resource Definitions

Tool: get_word_stats

• Description: Analyzes a string to return word count, character count, and complexity.
• Input Schema:

  {
    "text": "string"
  }
  

• Deterministic Logic: Pure Python code; no AI calls inside.

Resource: mcp://config/info

• Description: A static, read-only JSON data source containing system metadata.
• Data: Versioning, author (Rashmi Priya), and system status.

---

4. Setup and Execution

Prerequisites

• Python 3.10+
• Google Generative AI SDK

  python -m pip install -U google-generativeai
  

Run Commands

1. Deterministic Test (The "Remove the LLM" Test): Proves the server runs successfully without the LLM.

   python test_server.py
   

2. Full System Run: Starts the bridge between the LLM and the Server logic.

   python client.py
   

---

5. Engineering Maturity Features

• Defensive Validation: The server validates input types (e.g., ensuring text is a string) and returns structured JSON error messages if the LLM hallucinates arguments.
• Boundary Logging: Full transparency of data moving between the Client and Server using INFO: and SERVER_LOG: markers.
• Decoupled Logic: The server logic is completely usable without an LLM interface.

---

Developed by: Rashmi Priya (24BCE10054) Round: AI Club’s Tech Team Task Round