Integrations and MCP in AI-School
With integrations, the AI can use information from services such as Google Workspace, Microsoft 365, and an MCP server. An integration that you enable in Preferences → Tools becomes part of your standard toolset. This means you don’t have to start a separate integration chat for every question.
What types of integrations are there?
| Type | Examples | Web | Mobile | Desktop |
|---|---|---|---|---|
| Built-in integrations | Gmail, Google Calendar, Google Drive, Outlook, Teams, and SharePoint | Yes | Yes | Yes |
| Remote MCP | A registered Streamable HTTP server | Yes | Yes | Yes |
| Local MCP | Microsoft 365, Google Workspace, Nextcloud, and Memory via stdio | No | No | Yes |
The environment determines which integrations are available. You only see options you have access to.
Google Workspace and Microsoft 365
An administrator first makes the Google or Microsoft suite available. Then you can enable the desired integrations in Preferences → Tools and grant permission once.
- Google Workspace can provide Gmail, Google Calendar, and Google Drive.
- Microsoft 365 can provide Outlook Mail, Outlook Calendar, Teams, and SharePoint.
- Access tokens and refresh tokens are encrypted and stored server-side only.
- You can revoke permission later or reconnect the integration.
The integrations are executed by AI-School and are not tied to a single AI provider. You can use them with OpenAI, Claude, Gemini, and Mistral, as long as the chosen model supports tool usage. With Gemini 3 and newer, suitable Google-native tools and AI-School tools can be combined in the same context; older Gemini models use only the native toolset when native tools are active.
Using a remote MCP server
Remote MCP servers preferably use Streamable HTTP. The endpoint must:
- be a public HTTPS URL;
- include the full MCP path, usually
/mcp; - be reachable without redirects;
- support valid MCP initialization;
- be outside local, private, and metadata networks.
AI-School verifies a server before it becomes active. Chats then only receive a registered connectorId; URLs and secrets are not stored in the chat configuration.
Enabling integrations
- Open your profile menu and go to Preferences.
- Open Tools or Integrations.
- Enable an available connector or MCP server.
- Complete the sign-in when prompted.
- Start a regular chat and ask a specific question about the linked source.
In the chat settings, you see which integrations are available for that chat. The AI chooses a suitable tool when needed.
Desktop: local integrations
The desktop app uses the same built-in integrations for Google and Microsoft as web and mobile. Additionally, it can install local stdio-MCP servers, for example for Nextcloud, Memory, and other self-chosen services. Local file access is built-in and does not require an MCP server.
Read more at Local integrations in the desktop app.
Privacy and control
- A connector requests only the scopes needed for the chosen functions.
- Server-side credentials are not readable from the browser or mobile app.
- Local desktop secrets are stored encrypted on the device.
- An administrator determines which suites and environment servers are available.
- You decide which available integrations are enabled in your standard toolset.
- Actions with write permissions can create, modify, or delete data; always check the requested permissions carefully.
