How to Connect MCP Servers to Open WebUI

What is Open WebUI?

Open WebUI is a self-hosted web interface for running AI models locally via Ollama (and other backends). It supports tool calling through MCP, which means your local models can access external tools and APIs during conversations.

Why connect MCP servers to Open WebUI?

Local models are great at conversation but limited to their training data. Connecting MCP servers gives them live access to your business tools: check orders, query dashboards, look up tickets, and pull real-time data without sending anything to external AI providers.

Prerequisites

• Open WebUI running (typically at localhost:8080)
• Ollama with a tool-calling capable model (qwen2.5:32b, llama3.1:70b, or mistral-large)
• A DataFaucet account (free tier works)

Step 1: Create an MCP server with DataFaucet

1. Go to DataFaucet
2. Enter the URL of any web app you want your model to access
3. Browse the app while DataFaucet captures the API surface
4. Click Deploy to get your hosted MCP endpoint

Your endpoint: https://datafaucet.dev/mcp/your-server-id

Step 2: Add the MCP server in Open WebUI

Open WebUI supports MCP tool servers through its settings:

1. Open Settings (gear icon) in Open WebUI
2. Navigate to Tools or Connections section
3. Add a new MCP server connection
4. Enter your DataFaucet endpoint URL
5. Save and reload

Step 3: Select a capable model

Not all local models handle tool calling well. Recommended models for reliable MCP tool use:

• qwen2.5:32b or higher (most reliable for structured tool calls)
• llama3.1:70b (strong tool calling, needs 48GB+ VRAM)
• mistral-large (good balance of size and capability)
• command-r-plus (solid at multi-step tool workflows)

Smaller models (7B-13B) often struggle with complex tool schemas.

Step 4: Test the connection

Start a new chat and ask something that requires your connected tool:

• "What's the status of ticket PROJ-123?" (if connected to Linear/Jira)
• "Show me today's revenue" (if connected to Stripe/analytics)
• "Who signed up in the last hour?" (if connected to your auth dashboard)

The model will discover available tools from the MCP server and call them as needed.

Tips for best results

• Start with one MCP server before adding multiple
• Use larger models (32B+) for reliable tool calling
• Keep tool descriptions clear in your DataFaucet server (the model reads these)
• Set system prompts that tell the model it has tool access

Multiple MCP servers

You can connect multiple DataFaucet servers to Open WebUI simultaneously. Each one provides a different set of tools. Your model picks which to use based on the conversation.

Privacy advantage

This entire setup runs locally except for the MCP server calls. Your prompts and model inference stay on your machine. Only specific tool requests go to DataFaucet endpoints, which return structured data without seeing your full conversation.

Getting started

Create your first three MCP servers free at DataFaucet. Pair them with Open WebUI and a capable Ollama model for fully local AI with real tool access.

For more on local models and MCP, see our DataFaucet vs Ollama comparison.