# Ollama MCP Server Demo

A beginner-friendly project that demonstrates how to build a simple Model Context Protocol (MCP) server using Python and connect it to a local Ollama model.

## Overview

This project shows:

- How to interact with a local Ollama model using the REST API
- The difference between Ollama's Generate API and Chat API
- How to expose an Ollama model as an MCP tool
- How to configure and run an MCP server in VS Code

## Project Structure

```text
.
├── chat.py
├── generate.py
├── ollama_mcp_server.py
├── test_ollama.py
└── .vscode
    ├── mcp.json
    └── session.py
```

## Architecture

```text
MCP Client (VS Code)
        │
        ▼
Python MCP Server
        │
        ▼
Ollama API
        │
        ▼
gpt-oss:20b-cloud
```

## Prerequisites

- Python 3.10+
- Ollama installed and running
- A downloaded Ollama model (e.g. `gpt-oss:20b-cloud`)

## Installation

Clone the repository:

```bash
git clone <your-repository-url>
cd ollama-mcp-server-demo
```

Install dependencies:

```bash
pip install -r requirements.txt
```

## Running Ollama

Start Ollama:

```bash
ollama serve
```

Verify available models:

```bash
ollama list
```

## Testing Ollama

Run:

```bash
python test_ollama.py
```

If everything is working, the model will generate a response in the terminal.

## Running the MCP Server

Start the MCP server:

```bash
python ollama_mcp_server.py
```

The server exposes the following MCP tool:

```text
ask_ollama(prompt: str)
```

Example:

```text
Prompt: Explain MCP

Response: Model Context Protocol (MCP) is...
```

## Learning Objectives

This project helped me understand:

- MCP fundamentals
- MCP tools and servers
- Ollama APIs
- Local LLM integration
- VS Code MCP configuration
- Python-based AI tooling

## Future Improvements

- Add multiple MCP tools
- Support chat history
- Add document search (RAG)
- Connect to SharePoint or Microsoft Graph
- Add logging and error handling

## Author

Rohit Patil