DeepSeek runs on your own hardware with Ollama or LM Studio. Ollama alone does not connect to MCP servers: go through an interface. The most common is Open WebUI (version 0.6.31 or later, administrator account):
- In your Mailpro account, on the AI Assistants (MCP) page, copy the server address.
- In Open WebUI, open Settings › Admin › Integrations, then under External Tool Servers click + Add Connection.
- Type: MCP (Streamable HTTP). Paste the address into Server URL.
- Auth: OAuth 2.1, then click Register Client and Save.
- In a conversation, choose your DeepSeek model, then + › Tools and turn on Mailpro: the Mailpro sign-in page opens the first time.
- In Mailpro, on the AI Assistants page, set how it sends: with approval or direct, within a daily cap you set.
Recommended model: DeepSeek-R1, in a version that supports tools. A model that is too small handles tool calling poorly.
For LM Studio or LibreChat, and to set WEBUI_SECRET_KEY: the self-hosted AI guide. All about DeepSeek: DeepSeek and Mailpro.
Learn more
- Self-hosted AI: send your emails with Llama, Qwen or Mistral without your data leaving
- Definition: Large language model (LLM)
- The Mailpro MCP server
Mailpro and AI assistants
Your AI assistant, connected to Swiss email marketing
Connect Claude, ChatGPT, Mistral or your own model to Mailpro in a few minutes. You decide what it may send: approve each send, set a daily limit or suspend all sends at any time.