An AI Bridge to Sovereign Storage
We recently shipped v1.0.0 of the oCIS MCP Server.
This is the first stable release of an open source Model Context Protocol server that exposes ownCloud Infinite Scale as 80 tools, along with resources, and prompts for any MCP-compatible AI assistant — Claude Desktop, Claude Code, Ollama, and whatever comes next. Apache 2.0 licensed. Go binary. Works with any oCIS deployment.
What MCP Is and Why It Matters Here
The Model Context Protocol is an open standard developed by Anthropic that defines how AI assistants connect to external systems. Instead of writing custom integrations for every product, you build an MCP server once and every MCP-compatible client can use it. Think of it as an open protocol for AI tool access — the same principle behind why we built oCIS on WebDAV and OCM rather than proprietary APIs.
The oCIS MCP Server is the bridge between AI assistants and your sovereign file infrastructure. Once connected, your AI can manage users, create spaces, search files, handle shares, run federated OCM operations, and trigger multi-step workflows — all through natural language, all authenticated through your own oCIS instance, all with your data staying exactly where it is.
That last part is the point. The AI works on your infrastructure, not someone else’s. The files don’t leave your server. The queries go through your authentication. What the AI sees is what you’ve authorised it to read, and with a local model like Ollama, that content never leaves your infrastructure at all. This is what sovereign AI tooling looks like.
What’s in v1.0.0
80 tools across 13 categories:
- Users (6): List, get, create, update, delete users. Check the current authenticated user.
- Groups (7): Full group management — list, get, create, update, delete, add and remove members.
- Spaces / Drives (14): The complete Spaces API — list, get, create, update, disable, delete, restore; invite members, create links, manage permissions, empty trashbins, set space image and README.
- Files & Folders (14): List, upload, download, move, copy, delete. File versioning (list versions, restore). Tag and untag resources. Get resource metadata and retrieve by ID.
- Shares (11): Create shares and public links. List, update, and delete shares. Manage expiration. List shares sent and received. Accept and reject incoming shares. Query sharing roles.
- Search (2): Full-text search and tag-based search across your oCIS instance.
- Notifications (2): List and dismiss notifications.
- Settings / Roles (3): List available roles, assign roles to users, list role assignments.
- App Tokens (3): Create, list, and delete app tokens for service accounts and API access.
- Admin (4): Health check, version info, capabilities, and instance configuration.
- Education (5): Education-specific user and school management. Built for education-based deployments.
- OCM / Federated Sharing (4): Open Cloud Mesh operations — list federation providers, create and list OCM shares, view received federated shares.
- Multi-Step Workflows (5): Compound operations that combine multiple tools into single natural-language actions.
Two Authentication Methods
OIDC (OpenID Connect): Uses your existing identity provider. Works with Keycloak, LibreGraph Connect, and any OIDC-compliant IdP.
App Token: For service accounts and automated workflows. Generate a token in oCIS, configure the server with it, done.
How to Get It
Follow the Getting Started guide at github.com/owncloud/ocis-mcp-server/blob/main/GETTING_STARTED.md, which covers Claude Desktop, Ollama, on Mac, Windows, and Linux with a setup script that detects your system and configures everything.
- Pre-built binary: signed release binary, no Go toolchain required. GPG key 90CCF130D75586F0. Verify it.
- Docker: single docker run with environment variables.
- Build from source: git clone, make build. Requires Go 1.21+.
Why We Built This
MCP is the emerging standard for AI tool integration. oCIS is already built on open protocols — WebDAV, OCM, OIDC, LibreGraph. Adding an MCP server is the logical next step: one more open protocol, one more integration surface, one that happens to be the protocol most major AI assistants are now adopting.
The alternative is waiting for someone else to build a proprietary oCIS integration into a proprietary AI platform on proprietary infrastructure. That’s the opposite of what ownCloud is for.
We published an AI-Assisted Contribution Policy in May 2026 that says: great ideas should not be blocked by someone’s programming seniority.
The oCIS MCP Server is the operational expression of that principle.
Contributions welcome.
Links:
MCP Server v1.0.0 release
oCIS on GitHub
owncloud.dev
