Set up Getwello with Ollama

This guide runs the Getwello MCP server alongside a local Ollama model and an MCP-capable frontend. Your prompts and the model's responses stay on your machine; the only thing crossing the network is the Getwello API call itself.

What you need

• A machine with at least 8GB unified memory (Apple Silicon) or a recent GPU (Windows/Linux).
• Ollama installed (ollama.com).
• An MCP-capable frontend: Open WebUI is the most polished as of 2026.
• A Getwello Coordinator account with active subscription.

Step 1: Install and pick a model

brew install ollama  # macOS
ollama pull llama3.1
ollama run llama3.1

Llama 3.1 8B is a solid baseline for tool-calling. Qwen 2.5 7B and Mistral Nemo work too. Smaller (3B-class) models hallucinate tool calls; avoid for production.

Step 2: Install Open WebUI

Easiest is Docker:

docker run -d -p 3000:8080 \
  -v open-webui:/app/backend/data \
  --name open-webui ghcr.io/open-webui/open-webui:main

Open http://localhost:3000, finish the first-run setup.

Step 3: Add the Getwello MCP server to Open WebUI

Settings → Tools → Add MCP server. Paste the same npx config as Claude Desktop:

{
  "command": "npx",
  "args": ["-y", "@getwello-app/mcp-server"],
  "env": {
    "GETWELLO_API_KEY": "gw_live_your_real_key"
  }
}

Step 4: Try a query

Pick the Llama model, start a chat. Sixteen Getwello tools are available; the model picks the right one based on your question:

• “How is Mum doing this week?” — pulls stats + check-ins, summarises
• “Has she checked in today?” — calls get_check_in_today
• “Are there gap days next week?” — calls list_gap_days
• “Schedule me to visit Tuesday at 2pm.” — calls schedule_visit

All locally: the model never sends prompts off your machine, and the only network call is the Getwello API itself.

Troubleshooting

Model doesn't call the tool: use a model trained for tool-calling. Llama 3.1, Qwen 2.5, Mistral Nemo. Smaller models lack the training.

Slow responses: 7-8B models are slow on integrated GPUs. Try a quantised variant (Q4_K_M) or a smaller model.

Data privacy concern: the only network call is the Getwello API itself. The model never sees your prompts on any server but yours.