Tools
Via the Tools tile in the admin environment, you configure per AI provider which tools and integrations are available. This determines which additional capabilities the AI can use in a conversation, such as searching the internet, creating images, doing calculations, planning routes, or recording information.
The page opens on MCP. After that, the provider tabs from the model settings follow. If a provider is disabled there, the corresponding tool tab will not appear.
Microsoft organization app and write actions
Do you want to offer Microsoft 365 through your own organization app? Follow Set up Microsoft 365. There you will find the required Entra app, Graph permissions, organization consent, Identity Provider, and redirect URLs for AI-School, AI-Corporate, and AI-Public.
Under MCP → Suite shortcuts → Microsoft 365, you make the integrations available. If the suite is turned off, the underlying rows are hidden. In the Write actions column, you choose per integration Drafts, Send, Write, or Post; only approved actions can be performed. Click Save. Users only enable or disable the integrations that are available to them in Preferences.
Page structure
The page is divided into tabs per provider, plus a separate MCP tab:
| Tab | Meaning |
|---|---|
| MCP | Manage suites and MCP servers for the environment |
| OpenAI | Tool settings for OpenAI models |
| Claude | Tool settings for Claude models |
| Google (Gemini) | Tool settings for Gemini models |
| European AI | Tool settings for European models such as Mistral |
Per tab, depending on the provider, you manage:
- Native tools (built-in provider tools)
- Universal tools (tools from AI-Public)
On the MCP tab, you manage MCP servers at environment level. These are separate from the provider-specific tool settings.
The explanation for an individual tool is shown in the info button next to the tool name. This keeps the page compact while still allowing administrators to view extra context for each tool.
Google and Microsoft connectors
Under MCP → Suite shortcuts, you can make Google Workspace and Microsoft 365 available or unavailable as a group. This manages the provider-independent AI-Public integrations for the environment; an administrator does not sign in on behalf of all users here.
Users enable available connectors personally via Preferences → Tools and grant consent there once. Their tokens are stored server-side and remain usable in new chats and on other devices.
Examples of available MCP connectors:
- Google Calendar
- Gmail
- Google Drive
- Microsoft Teams
- Outlook Calendar
- Outlook Mail
The Google and Microsoft tools are run by AI-Public and are not tied to OpenAI. Users can use them with OpenAI, Claude, Gemini, and Mistral, as far as the chosen model supports tool use.
Google Calendar, Gmail, and Google Drive: own Google app
Google Calendar, Gmail, and Google Drive require read permissions that Google classifies as sensitive or restricted. These are only available through an internal Google app from your own organization. Without your own app, they appear greyed out and the Google Workspace suite in MCP → Suite shortcuts is unavailable. Signing in with Google and uploading files to a user’s own Google Drive still work through the central AI-Public app.
An internal app (user type Internal) does not need to be verified by Google and does not require a security assessment. Only accounts within your Google Workspace can use it to sign in and connect. It works like the own Entra app for Microsoft:
- Create an OAuth consent screen in Google Cloud in a project belonging to your own organization, with user type Internal. Add the scopes
openid,email,profile,calendar.events.readonly,drive.file,drive.readonly, andgmail.readonly, and enable the Gmail API, Google Drive API, Google Calendar API, and Google Picker API. - Create an OAuth client of type Web application with these two authorized redirect URIs:
https://ai-public-pro.firebaseapp.com/__/auth/handler(sign in with Google)https://europe-west1-ai-public-pro.cloudfunctions.net/completeConnectorOAuth(connectors)
- Provide the client ID and client secret through the agreed secure channel. The platform administrator will configure them on the Google provider of your environment tenant in Google Cloud Identity Platform, enable the Google sign-in method under Admin → Environment → Tenant, and fill in the Google Workspace domain, for example
organisatie.nl. - After that, you can enable Google Calendar, Gmail, and Google Drive. Users with a previous Google connection reconnect their Google account once.
Signing in with Google then also uses the own app and shows only accounts from the configured domain.
Tasks: Microsoft To Do and Google Tasks
The integrations list also includes Microsoft To Do and Google Tasks. These are not AI tools: they keep users’ My tasks in sync two-way with a list in that service. You enable the service with the switch in the row; users then connect it themselves, in Knowledge base → Tasks or under Preferences → Integrations. One service can be active per user at a time.
- Microsoft To Do requires the delegated Graph permission
Tasks.ReadWritein your Entra app (see Set up Microsoft 365) and a Microsoft account with an Exchange Online mailbox. Without a mailbox, the app reports that a tasks list cannot be created. When connecting, Microsoft only asks for the new permission; permissions granted earlier for, for example, email or files are not requested again. - Google Tasks works only with your organization’s own Google app, as described above. Without that app, the row appears gray with the message that a custom Google app is required. In the consent screen, add the scope
https://www.googleapis.com/auth/tasksand enable the Google Tasks API. Users then reconnect Google once.
For users, the explanation is under Tasks.
Environment MCP servers
On the MCP tab, you can first open a store with suitable presets via Add MCP server. The web app only shows servers that can work via a public Streamable HTTP connection. Via Add your own MCP server, you open the manual form.
Use the full HTTPS endpoint, including the MCP path such as /mcp. The backend checks DNS, private addresses, redirects, and the MCP handshake. After registration, chats only use a connectorId; endpoints and secrets are not stored in the chat configuration.
An MCP server configured here appears on the dashboard under Tools for users who are allowed to see MCP connectors. They see these environment servers alongside their own personal MCP servers and can start a chat with them directly.
Users cannot edit or remove environment servers themselves. They can, however, enable or disable an available server personally. Management is done centrally via Admin environment → Tools → MCP.
As an administrator, you can edit, enable or disable, and delete a configured server in this list. Users who have enabled the server keep it enabled. If you change the endpoint or authentication method, re-enter the secret and users will sign in again via OAuth. A disabled server is no longer available to users.
Environment servers are only visible to users who have access to MCP connectors. This access is determined by the environment limits and any individual settings for employees.
Native tools
Native tools are built-in tools from the provider itself. They are not run by AI-Public, but are offered by the provider with the model.
| Tool | Provider | Description |
|---|---|---|
| OpenAI Web Search | OpenAI | Searches current information via OpenAI’s built-in web search |
| OpenAI Code Interpreter | OpenAI | Runs Python code for data analysis, calculations, and visualizations |
| Claude Web Search | Claude | Searches current information via Claude’s built-in search |
| Claude Web Fetch | Claude | Retrieves content from web pages and PDF documents |
| Claude Code Execution | Claude | Runs Bash and Python code in a secure sandbox |
| Gemini Google Search | Searches current information via Gemini’s built-in Google Search | |
| Gemini Code Execution | Runs code for calculations and data analysis | |
| Gemini URL Context | Retrieves content from URLs for extra context | |
| Gemini Google Maps | Uses Google Maps for location information, routes, and places |
Gemini 3 and newer models can combine suitable Google-native tools with universal tools and Google or Microsoft connectors from AI-Public. For older Gemini models, universal tools are disabled in a context with active native tools to prevent incompatible tool combinations.
Code execution tools, such as OpenAI Code Interpreter, Claude Code Execution, and Gemini Code Execution, are off by default. Enable them deliberately when code execution, calculations, data analysis, or file editing is needed.
Universal tools
Universal tools are available for the providers where they technically fit and are offered by AI-Public itself. For each provider, you choose what happens per tool.
| Tool | Description |
|---|---|
| Create image | Generates images based on a text description |
| Create practice questions | Automatically creates test or practice questions about a text or topic |
| Search internet | Searches for current information on the internet |
| Search images | Searches for images on the internet based on a search term |
| Get weather | Retrieves the current weather forecast for a location |
| Check AI content | Analyzes whether a text may have been generated by AI |
| Search Wikipedia | Searches Wikipedia for encyclopedic information |
| Google Routes | Plans routes and calculates travel times |
| Google Places | Searches for information about locations and businesses |
| Wolfram Alpha | Performs complex calculations and scientific analyses |
| Recording | Saves important information as an editable note in the chat |
Some internal tools, such as chat_history_search, do not appear in this overview. They are automatically available to the assistant and follow the same environment, user permissions, and retention period. Administrators cannot manage them separately through the tool settings.
Setting per tool
For each tool, you choose from four values:
| Setting | Meaning |
|---|---|
| Automatic | The AI decides for itself whether the tool is useful for the question |
| On | The tool is always enabled in the relevant context |
| Off | The tool is disabled and cannot be used |
| Not visible | The tool disappears completely from the user interface and from personal preferences |
The Automatic setting is the recommended choice in most cases. Choose On only when a tool is almost always needed in that context.
In their preferences, users can only set personal exceptions for visible tools. They cannot override the environment setting for tools that are set to Not visible.
Reset and save
Via Restore default values, you reset the tool settings of a provider to the default settings of AI-Public. Changes are saved with the Save button in the app toolbar.
See also
- System prompts – steer AI behavior via system instructions
- Models – configure available providers and models
- Tools and connectors (user guide) – explanation of the tools from the user perspective