Sandesh2905/genai-inventory-mcp-server

This project demonstrates a backend system that enables users to manage inventory using natural language. It consists of two services:

1. Inventory Service --> (Inventory_services/application.py)
   A RESTful Flask API to add, sell, and check item stock.
2. Model Control Plane (MCP) --> (MCPServer/mcp.py)
   Converts natural language queries into structured API calls using OpenRouter’s LLMs.

My Approach: I have cratead a Flask for an invetory and the exposed the contents to the endpoint /inventory. For my idea of the MCP, I decided to parse the NLP query to a LLM (OpenRouter in my project) and set instruction for the LLM to convert the query to a post or a get call based on query, for the propmt template I added support queries as mentiond in the OpenAI docs to make my prompt a bit more strict, I have also added the format for the curl commands in the LLM so that tLLM understand that the query need to be converted to a curl command and reutrn method_for_query, item, change. once the LLM returns the Query in the form of {item,change,method} based on the method I will perform POST or GET call on the appropriate items and Update the inventory.

**** Important **** Not all models generate support the code, the few I used and are working are :

1. "moonshotai/kimi-k2:free" (currently using)
2. nvidia/llama-3.3-nemotron-super-49b-v1:free (used for 1 day and the free version was removed)

Steps to Execute the project (I have used powershell, if using linux, please change the terminal commands)

Step 1: Clone the Git Repo

Step 2: Activate the Virtual Environment $ . .\venv\Scripts\Activate.ps1

Step 3: Install requirements $ pip install -r requirements.txt

Step 4: Set OpenRouter API Key: $ OPENROUTER_API_KEY=openrouter_key_here

Step 5: Start the Inventory_services $ cd Inventory_services $ python application.py

Step 6: Start the MPC server $ cd MCPServer $ python mcp.py

Step 7: Open a new shell and runn the following command :

Example commands to test the project:

for linux: 
$  curl -X POST http://localhost:6000/query \
 -H "Content-Type: application/json" \
 -d '{"query": "Add 20 iPhones"}'

for powershell (without NLP):
$ Invoke-RestMethod -Uri "http://localhost:5000/inventory" `
    -Method POST `
    -Headers @{ "Content-Type" = "application/json" } `
    -Body '{"item": "tshirts", "change": -3}'

with NLP:
Invoke-RestMethod -Uri "http://localhost:6000/query" -Method POST -Headers @{ "Content-Type" = "application/json" } -Body '{"query": "remove 2 tshirts"}'

Sandesh Gowda Shirgavara Ganesha
https://www.linkedin.com/in/sandesh-gowda/