AIquila — Nextcloud MCP Server

AIquila

Private AI for your Nextcloud — your data, your server, your choice of AI provider

What is AIquila?

AIquila brings AI to your self-hosted Nextcloud. Instead of keeping your files, notes, tasks and recipes locked inside Nextcloud — or copying them by hand into a chat window — AIquila lets an AI assistant read and write your Nextcloud data directly. Your data stays on your own server, and you decide which AI provider sees it: a hosted API, or a model running on your own hardware that sends nothing to anyone.

How it works

AIquila has three components that can be used independently or together:

MCP Server — A Model Context Protocol server that gives any MCP-compatible AI assistant secure access to your Nextcloud. 342 tools across 44 categories: browse and manage files, keep calendars, tasks and contacts in sync, work in Talk, Deck and Mail, organise photos, notes and bookmarks, and run Coworkers and occ administration. It has no model of its own — it exposes Nextcloud to whichever MCP client you connect.

Nextcloud App — A native Nextcloud application that brings AI directly into the Nextcloud UI. Chat about your documents, summarise and rewrite text, run Coworkers (saved, repeatable AI jobs), and scope a conversation to a Project so the assistant only sees the folders you choose. Admins pick which providers are available, and may restrict them per user or group.

Hetzner Deployment — A single-command provisioning tool (aiquila-hetzner) that stands up a production-ready AIquila server on Hetzner Cloud, complete with Traefik reverse proxy, CrowdSec intrusion prevention, TLS, and optional monitoring.

AI providers

The Nextcloud app supports five providers for chat. Each user can pick a provider and model per conversation, within what the admin allows.

Provider choice applies to chat, the per-conversation model picker and the Nextcloud Assistant / TaskProcessing integrations, which all run on the provider the user has selected. Assistant actions that send images ("Analyze images", "Extract text from image") need a vision-capable provider, and report which of the available ones qualify when the selected provider is not.

Getting Started

Pick the path that fits your setup:

Shortened. The full README is on GitHub.