Many companies run AI in-house to keep control of their data. An open model can't do anything on its own: it acts through an MCP server. The Mailpro one gives it your contacts, your templates and the ability to send emails and SMS.
A self-hosted AI has two parts. The engine runs the model: Ollama or LM Studio. The interface chats with it and gives it tools: Open WebUI, LibreChat or LM Studio itself.
It is the interface that connects to the Mailpro MCP server. Open WebUI does it natively since its fall 2025 releases, and so do LibreChat and LM Studio, all with OAuth sign-in.
Your questions and the AI's answers stay on your machines. Only the calls to Mailpro go out over the Internet, to your Mailpro account hosted in Switzerland.
| Engines | Ollama, LM Studio |
|---|---|
| Interfaces with MCP | Open WebUI, LibreChat, LM Studio |
| Sign-in | OAuth 2.1 with automatic registration |
| What goes out | Only the calls to Mailpro |
| Models | Qwen, Llama, DeepSeek, Gemma, Mistral… |
Your internal documents, your questions and the drafts stay in your network.
No per-user subscription to an online assistant: the model runs on your hardware.
A small model for simple tasks, a large one for analysis: switch whenever you like.
Approval, daily cap, log and suspension, just like with a major assistant.
Open WebUI (admin): Admin Settings › Integrations › External Tool Servers › + Add Connection, type MCP (Streamable HTTP), OAuth 2.1 authentication.
Enter the server address shown in your Mailpro account, on the AI Assistants (MCP) page, then sign in to Mailpro and accept the permissions; in a conversation, turn on the tool from the + menu.
LibreChat: the admin declares the server in librechat.yaml under mcpServers, with OAuth sign-in and no credentials to enter.
LM Studio: in mcp.json, add the server with its url; LM Studio opens the OAuth sign-in in the browser.
Ollama runs the model but does not connect to MCP servers by itself: use it with Open WebUI or LibreChat.
Choose a model that supports function calling; a model that is too small ignores the tools or gets the arguments wrong.
Set a fixed WEBUI_SECRET_KEY variable, otherwise OAuth connections are lost on every restart.
The model and the conversations stay with you; the calls to Mailpro go out over HTTPS to your account.
On the Mailpro side, the rules are the same for every assistant: by default, each send waits for your approval, or goes out directly within a daily cap; a log keeps track of everything, and one click suspends all sends. The details.
No: Ollama runs the model. To use Mailpro, go through an interface that supports MCP servers, such as Open WebUI or LibreChat.
A model that supports function calling, large enough: Qwen, Llama, Mistral, DeepSeek and Gemma all have suitable versions.
The conversations do. The calls to Mailpro go out to your Mailpro account, hosted in Switzerland.
All questions: AI assistants FAQ.
Create a free Mailpro account, connect your assistant from the AI Assistants page of your account, and stay in control of every send.
See also: the Mailpro MCP server · how to choose an MCP server · what is a language model · Qwen · Llama · DeepSeek · Gemma · Mistral
