AI-Assisted CRM Using MCP Server

A full-stack CRM application integrated with an MCP (Model Context Protocol) server, built for learning how real-world AI agents communicate with backend services. The project includes a built-in chatbot that lets you perform CRM operations — creating contacts, managing deals, updating leads — using natural language prompts.

I built this using FastAPI, which made it straightforward to expose every endpoint as both a REST API and an MCP tool simultaneously. The goal was to demonstrate how modern applications can be designed to serve both human users and AI agents from a single codebase.

This project is purely for educational purposes — to understand how to build your own MCP server, how to structure tools that AI agents can understand, and how AI agents communicate with MCP servers in a real-world context.

Why I Built This

AI agents are everywhere now. Most developers know how to build APIs for users, but building APIs that AI agents can reliably understand and use is a different skill. This project is my attempt to bridge that gap with a real, working example.

What I handled in this project:

• Designed the system architecture and layered structure

• Built the MCP server on top of FastAPI endpoints

• Created the LLM client and AI agent

• Implemented Redis-based cache memory for faster agent-to-MCP communication and to avoid redundant API calls

• Secured all MCP access with JWT authentication via request headers

• Enforced role-based access control for both the REST API and MCP tools

Related MCP server: Zoho CRM MCP Server

Architecture Overview

CRM + MCP System Architecture

Layered Architecture

The LLM agent reads the user prompt, decides which MCP tool to call, executes the corresponding CRM API, and returns the result — all automatically.

Project Structure

AI-Assisted-CRM-Using-MCP-Server
├─ app
│  ├─ chatbot
│  │  ├─ cache_service.py
│  │  ├─ handle_error.py
│  │  ├─ llm_client.py
│  │  └─ redis_client.py
│  │
│  ├─ core
│  │  ├─ cache_invalidator.py
│  │  ├─ config.py
│  │  └─ security.py
│  │
│  ├─ dependencies
│  │  ├─ auth.py
│  │  └─ permission.py
│  │
│  ├─ models
│  │  ├─ contacts.py
│  │  ├─ deals.py
│  │  ├─ leads.py
│  │  ├─ tasks.py
│  │  └─ users.py
│  │
│  ├─ routes
│  │  ├─ routers
│  │  │  ├─ auth.py
│  │  │  ├─ chat.py
│  │  │  ├─ contacts.py
│  │  │  ├─ dashboard.py
│  │  │  ├─ deals.py
│  │  │  ├─ leads.py
│  │  │  └─ tasks.py
│  │  └─ routes.py
│  │
│  ├─ schemas
│  │  ├─ chat.py
│  │  ├─ contacts.py
│  │  ├─ deals.py
│  │  ├─ leads.py
│  │  ├─ tasks.py
│  │  └─ users.py
│  │
│  ├─ static
│  │  ├─ css
│  │  └─ js
│  │
│  ├─ templates
│  │  ├─ dashboard.html
│  │  ├─ contacts.html
│  │  ├─ deals.html
│  │  ├─ leads.html
│  │  ├─ tasks.html
│  │  ├─ login.html
│  │  └─ register.html
│  │
│  ├─ main.py
│  └─ db.py
│
├─ client.py
├─ run.py
├─ requirements.txt
└─ README.md

Installation

1. Clone the Repository

git clone https://github.com/yourusername/AI-Assisted-CRM-Using-MCP-Server.git
cd AI-Assisted-CRM-Using-MCP-Server

2. Install uv

I use uv for package management — it's significantly faster than pip.

pip install uv

3. Create a Virtual Environment

uv init
uv venv

Activate it:

Windows

.venv\Scripts\activate

Linux / macOS

source .venv/bin/activate

4. Install Dependencies

uv pip install -r requirements.txt

5. Configure Environment Variables

Create a .env file in the project root:

GROQ_API_KEY=your_groq_api_key
JWT_SECRET_KEY=your_secret_key

Running the Application

python run.py

The server will start at:

http://127.0.0.1:8000

API Documentation

FastAPI's interactive Swagger UI is available at:

http://127.0.0.1:8000/docs

MCP Endpoint

The MCP server is exposed at:

/llm/mcp

This is the endpoint that AI agents use to interact with the CRM backend as a set of callable MCP tools.

Example Chat Interaction

The chatbot supports natural language commands for creating, updating, and retrieving records. Delete operations and queries using raw IDs are intentionally not supported through the chat interface.

User prompt:

Create a new contact named John Doe with phone number 9876543210

What happens internally:

1. The AI agent interprets the intent from the prompt

2. It selects and calls the appropriate MCP tool

3. The tool executes the corresponding CRM API

4. The result is returned to the user in natural language

License

This project is open-source and available under the MIT License.