# MCP Server Creation and MCP client connection (Agent-Claude,cursor or, custom server calling and integration with local LLMs).

This project demonstrates how to create and connect an MCP (Model Context Protocol) weather server using FastMCP.

> 📖 **Want to learn more about MCP?** Check out our [Comprehensive MCP Guide](Readme_MCP.md) for detailed explanations, concepts, and best practices.

## Prerequisites

- Python 3.13+
- `uv` package manager installed

## Setup Instructions

### 1. Initialize the Project

Initialize a new project using `uv`:

```bash
uv init
```

### 2. Create Virtual Environment

Create a virtual environment for the project:

```bash
uv venv
```

### 3. Activate Virtual Environment

Activate the virtual environment:

**Windows:**
```bash
.venv\Scripts\activate
```

**Linux/Mac:**
```bash
source .venv/bin/activate
```

### 4. Install MCP CLI

Add the MCP CLI package to enable fast MCP CLI commands:

```bash
uv add "mcp[cli]"
```

## Running the MCP Server

### Development Mode

To run the MCP server in development mode:

```bash
uv run mcp dev server/weather.py
```

**Expected Output:**
```
Starting MCP inspector...

⚙️ Proxy server listening on localhost:6277
```

## Connecting to Claude Desktop

### Install Server to Claude

Add the weather server to Claude Desktop:

```bash
uv run mcp install server/weather.py
```

**Expected Output:**
```
Added server 'weather' to Claude config 

Successfully installed weather in Claude app
```

Once installed, you can ask Claude questions like:
- "What are the weather alerts in CA?"
- The MCP server will be called automatically and display the results.

### Manual Configuration for Cursor or Other Clients

For Cursor or other clients, you'll need to manually configure the server using the Claude Desktop config file format.

**Get the server configuration from:** `claude_desktop_config` file

**Example Configuration:**
```json
"weather": {
  "command": "C:\\Users\\gaura\\AppData\\Local\\Programs\\Python\\Python313\\Scripts\\uv.EXE",
  "args": [
    "run",
    "--with",
    "mcp[cli]",
    "mcp",
    "run",
    "C:\\Users\\gaura\\OneDrive\\Desktop\\AI projects\\mcp_project\\server\\weather.py"
  ]
}
```

**Note:** Update the paths in the configuration to match your system paths.

**Usage in Cursor:**
- Go to **MCP** → **mcp config.json**
- Add the configuration above

## Using MCP with Local LLMs (mcp-use)

`mcp-use` allows you to connect to MCP servers directly without an AI agent for programmatic tool access. This is useful for custom server calling and integration with local LLMs.

### Installation

Install the `mcp-use` package:

```bash
uv add mcp-use
```

### Configuration

Create a file named `weather.json` and add the following configuration to connect local LLMs to the MCP server:

```json
{
  "mcpServers": {
    "weather": {
      "command": "C:\\Users\\gaura\\AppData\\Local\\Programs\\Python\\Python313\\Scripts\\uv.EXE",
      "args": [
        "run",
        "--with",
        "mcp[cli]",
        "mcp",
        "run",
        "C:\\Users\\gaura\\OneDrive\\Desktop\\AI projects\\mcp_project\\server\\weather.py"
      ]
    }
  }
}
```

**Note:** Update the paths in the configuration to match your system paths.

## References

- **FastMCP**: https://github.com/jlowin/fastmcp
  - Create MCP servers and connect them to different clients like Claude, Cursor, etc.

- **mcp-use**: https://github.com/mcp-use/mcp-use
  - Connect to MCP servers directly without an AI agent for programmatic tool access (custom server calling and use)