Local AI with Ollama, WebUI & MCP on Windows
A self-hosted AI stack combining Ollama for running language models, Open WebUI for user-friendly chat interaction, and MCP for centralized model management—offering full control, privacy, and flexibility without relying on the cloud.
This sample project provides an MCP-based tool server for managing employee leave balance, applications, and history. It is exposed via OpenAPI using mcpo for easy integration with Open WebUI or other OpenAPI-compatible clients.
🚀 Features
- ✅ Check employee leave balance
- 📆 Apply for leave on specific dates
- 📜 View leave history
- 🙋 Personalized greeting functionality
📁 Project Structure
leave-manager/
├── main.py # MCP server logic for leave management
├── requirements.txt # Python dependencies for the MCP server
├── Dockerfile # Docker image configuration for the leave manager
├── docker-compose.yml # Docker Compose file to run leave manager and Open WebUI
└── README.md # Project documentation (this file)
📋 Prerequisites
- Windows 10 or later (required for Ollama)
- Docker Desktop for Windows (required for Open WebUI and MCP)
- Install from: Docker Desktop for Windows
🛠️ Workflow
- Install Ollama on Windows
- Pull the
deepseek-r1model - Clone the repository and navigate to the project directory
- Run the
docker-compose.ymlfile to launch services
Install Ollama
➤ Windows
-
Download the Installer:
- Visit Ollama Download and click Download for Windows to get
OllamaSetup.exe. - Alternatively, download from Ollama GitHub Releases.
- Visit Ollama Download and click Download for Windows to get
-
Run the Installer:
- Execute
OllamaSetup.exeand follow the installation prompts. - After installation, Ollama runs as a background service, accessible at: http://localhost:11434.
- Verify in your browser; you should see:
Ollama is running
- Execute
-
Start Ollama Server (if not already running):
ollama serve- Access the server at: http://localhost:11434.
Verify Installation
Check the installed version of Ollama:
ollama --version
Expected Output:
ollama version 0.7.1
Pull the deepseek-r1 Model
1. Pull the Default Model (7B):
Using PoweShell
ollama pull deepseek-r1
To Pull Specific Versions:
ollama run deepseek-r1:1.5b
ollama run deepseek-r1:671b
2. List Installed Models:
ollama list
Expected:
Expected Output:
NAME ID SIZE
deepseek-r1:latest xxxxxxxxxxxx X.X GB
4. Alternative Check via API:
curl http://localhost:11434/api/tags
Expected Output:
A JSON response listing installed models, including deepseek-r1:latest.
4. Test the API via PowerShell:
Invoke-RestMethod -Uri http://localhost:11434/api/generate -Method Post -Body '{"model": "deepseek-r1", "prompt": "Hello, world!", "stream": false}' -ContentType "application/json"
Expected Response: A JSON object containing the model's response to the "Hello, world!" prompt.
5. Run and Chat the Model via PowerShell:
ollama run deepseek-r1
- This opens an interactive chat session with the
deepseek-r1model. - Type
/byeand pressEnterto exit the chat session.
🐳 Run Open WebUI and MCP Server with Docker Compose
-
Clone the Repository:
git clone https://github.com/ahmad-act/Local-AI-with-Ollama-Open-WebUI-MCP-on-Windows.git cd Local-AI-with-Ollama-Open-WebUI-MCP-on-Windows -
To launch both the MCP tool and Open WebUI locally (on Docker Desktop):
docker-compose up --build
This will:
- Start the Leave Manager (MCP Server) tool on port
8000 - Launch Open WebUI at http://localhost:3000
🌐 Add MCP Tools to Open WebUI
The MCP tools are exposed via the OpenAPI specification at: http://localhost:8000/openapi.json.
- Open http://localhost:3000 in your browser.
- Click the Profile Icon and navigate to Settings.
- Select the Tools menu and click the Add (+) Button.
- Add a new tool by entering the URL: http://localhost:8000/.
💬 Example Prompts
Use these prompts in Open WebUI to interact with the Leave Manager tool:
- Check Leave Balance:
Check how many leave days are left for employee E001
- Apply for Leave:
Apply  - View Leave History:
What's the leave history of E001?
- Personalized Greeting:
Greet me as Alice
🛠️ Troubleshooting
- Ollama not running: Ensure the service is active (
ollama serve) and check http://localhost:11434. - Docker issues: Verify Docker Desktop is running and you have sufficient disk space.
- Model not found: Confirm the
deepseek-r1model is listed withollama list. - Port conflicts: Ensure ports
11434,3000, and8000are free.