custom-mcp-server

reddybharat/custom-mcp-server

3.1

If you are the rightful owner of custom-mcp-server and would like to certify it and/or have it hosted online, please leave a comment on the right or send an email to henry@mcphub.com.

Custom MCP server and tools, with ReAct agent

Custom MCP Server & Client

A demonstration of the Model Context Protocol (MCP) architecture with custom servers and a ReAct agent client. This project showcases how to build modular, tool-based AI systems using MCP.

What is MCP?

The Model Context Protocol (MCP) is a standard for connecting AI models to external tools and data sources. It enables:

  • Modular Architecture: Separate tools into independent servers
  • Tool Discovery: Automatic tool registration and discovery
  • Multiple Transports: Support for stdio, HTTP, and other communication methods
  • Type Safety: Strongly typed tool definitions with automatic validation

Architecture Overview

┌─────────────────┐    ┌─────────────────┐
│   ReAct Agent   │    │  Calculator     │
│   (Client)      │◄──►│  Server         │
│                 │    │  (stdio)        │
└─────────────────┘    └─────────────────┘
         │
         │ HTTP
         ▼
┌─────────────────┐
│   Weather       │
│   Server        │
│   (HTTP)        │
└─────────────────┘

Project Structure

custom-mcp-server/
├── client.py              # ReAct agent client
├── server/
│   ├── calculator.py      # Calculator MCP server (stdio)
│   └── weather.py         # Weather MCP server (HTTP)
├── requirements.txt       # Dependencies
└── pyproject.toml        # Project configuration

Key Components

1. MCP Servers

Calculator Server (server/calculator.py):

  • Uses stdio transport for direct communication
  • Provides basic math operations: add, subtract, multiply, divide
  • Demonstrates simple tool definition with FastMCP

Weather Server (server/weather.py):

  • Uses streamable-http transport for web communication
  • Integrates with external weather API
  • Shows async tool implementation

2. MCP Client

ReAct Agent (client.py):

  • Connects to multiple MCP servers simultaneously
  • Uses LangGraph's ReAct agent for tool orchestration
  • Demonstrates multi-server tool integration

Quick Start

  1. Install Dependencies:

    pip install -r requirements.txt

  2. Set Environment Variables:

    export GROQ_API_KEY="your-groq-api-key"
export WEATHER_API_KEY="your-weather-api-key"

  3. Run the Weather Server (in one terminal):

    python server/weather.py

  4. Run the Client (in another terminal):

    python client.py

MCP Learning Points

Server Development

  • FastMCP: Simplifies MCP server creation with decorators
  • Tool Definition: Use type hints and docstrings for automatic schema generation
  • Transport Selection: Choose between stdio, HTTP, or other transports based on use case

Client Development

  • Multi-Server Support: Connect to multiple MCP servers simultaneously
  • Tool Discovery: Automatic tool registration from connected servers
  • Agent Integration: Use with LangGraph, LangChain, or other agent frameworks

Best Practices

  • Type Safety: Always use type hints for better tool validation
  • Error Handling: Implement proper error handling in tools
  • Documentation: Write clear docstrings for tool descriptions
  • Transport Choice: Use stdio for local tools, HTTP for distributed systems