tools-only MCP Tool Integration
This module provides integration with MCP (Model Context Protocol) services for Wilmer's AI system. It allows LLMs to automatically discover and invoke tools from MCP servers mentioned in the system prompt.
Important Note on MCP:
Wilmer has no way to validate or control what the MCP server or the tools it exposes may do, so please be sure to use only trusted and secure MCP servers. Like a front end, Wilmer can only call the tools; what those tools do is the responsibility of the server exposing them. Take care to ensure that you know exactly what those tools will do before using them.
This functionality is still new, and may have bugs or other issues that still need to be worked out. Please exercise caution when using it.
Overview
The integration consists of two main Python modules:
mcp_tool_executor.py- Handles discovering available tools from MCP servers and executing tool callsmcp_workflow_integration.py- Integrates MCP tools with Wilmer's workflow system
Configuration
The following configurations have been set up:
- Workflow:
WilmerData/Public/Configs/Workflows/isevendays-openwebui-norouting-general-offline-mcp/MCPToolsWorkflow.json - Endpoint:
WilmerData/Public/Configs/Endpoints/isevendays-openwebui-norouting-general-offline-mcp/Worker-Endpoint.json - User:
WilmerData/Public/Configs/Users/isevendays-openwebui-norouting-general-offline-mcp.json
How It Works
- The system automatically detects MCP services mentioned in the system prompt.
- It discovers available tools from those services using the OpenAPI schemas.
- It updates the system prompt with tool definitions.
- When the LLM generates a tool call, it executes the tool and formats the results for the next response.
Using MCP Tools
To use MCP tools in your prompts, simply mention the MCP services you want to use in OpenWebUI System Prompt. For example:
You are a helpful AI assistant with access to the following MCP services: time, openweather.
When you need to get the current time or weather information, use the appropriate tool.
The system will automatically discover the available tools from the mentioned services and make them available to the LLM. The underlying code is generic and should be able to handle most any tool calling that follows the MCP standard.
MCPO Server
What is MCPO?
MCPO (MCP-to-OpenAPI) is a proxy server developed by the Open WebUI team that allows MCP tools to be exposed as standard OpenAPI/REST endpoints. This solves several challenges with raw MCP servers:
- 🔓 Raw MCP servers communicate over stdio, which lacks security features
- ❌ Raw MCP is incompatible with many tools and services
- 🧩 Raw MCP lacks standard features like documentation, authentication, and error handling
Configuration Options
The integration is configured to use the MCPO server running at http://localhost:8889 by default. You can change this URL in the User's config file. An MCPO_URL environment variable can also be set, which will be used as a fallback if nothing is set in the User config.
Setting Up Your Own MCPO Server
If you want to run your own MCPO server to expose MCP tools as REST APIs, you can use the mcpo tool from Open WebUI:
Install mcpo:
pip install mcpoRun an MCP service through MCPO:
mcpo --port 8889 -- your-mcp-service --your-service-args
For example, to run a time service:
mcpo --port 8889 -- mcp-server-time --local-timezone=America/New_York
This will start a proxy server at http://localhost:8889 that exposes your MCP service's capabilities as REST APIs with auto-generated OpenAPI documentation.
Benefits of Using MCPO
- ✅ Works instantly with OpenAPI tools, SDKs, and UIs
- 🛡 Adds security, stability, and scalability using trusted web standards
- 🧠 Auto-generates interactive docs for every tool
- 🔌 Uses pure HTTP—no sockets, no glue code, no surprises
Integration with Open WebUI
This MCP tool integration is designed to work with MCPO.
If you're using Open WebUI with this integration, you can take advantage of their MCPO implementation for a more robust and feature-rich experience.