Set up and configure a tool
Setting up a new tool
Section titled “Setting up a new tool”Create from a template (recommended)
Section titled “Create from a template (recommended)”- Go to Settings → System integrations
- Click on ”+ Tool” → “From template”
- Choose a template
What happens next depends on the type of template:
Skill templates (e.g. Microsoft 365, DHL Tracking, Google Workspace): The credentials dialog or the sign-in with the provider opens directly. After saving, access is tested automatically and you are done. You then manage the stored credentials as a connection under Settings → Security in the “Credentials” section.
MCP server templates (e.g. Notion): The tool is created and the editor opens. Follow the integration guide in the “Parameters” tab and complete the setup with “Test connection”.
Set up an MCP server manually
Section titled “Set up an MCP server manually”For MCP servers that are not available as a template:
- Click on ”+ Tool” → “Manual MCP server”
- The editor opens
- In the “General” tab: enter a name and, optionally, a description
- In the “Parameters” tab: enter the server URL
- Optional: define parameters that the server requires (e.g. API key, username)
- Click on “Test connection”
- On success, the available capabilities are detected automatically
Your own skills
Section titled “Your own skills”You no longer build custom automations via tool management but directly in the chat: in the Skills area in the sidebar. There you create skills in conversation, connect them to external services when needed and share them with your team. All details under Skills.
Configure a tool
Section titled “Configure a tool”In the tool overview, click on a tool to open the editor. The configuration is split across several tabs. Which tabs are visible depends on the type of tool: MCP servers show General, Parameters, Connections, Sharing, Capabilities and Audit. Skill integrations (from skill templates) get by with General, Connections and Audit; they are shared through their connection.
Tab: General
Section titled “Tab: General”- Tool enabled: turns the tool on or off. Disabled tools are not available to the AI.
- Name: display name of the tool (max. 100 characters)
- Description: brief explanation of what the tool is for (max. 500 characters)
- System instructions: here you can tell the AI when and how to use this tool (max. 2,000 characters). Example: “Only use this tool when the user explicitly asks for Databricks data.”
- Delete tool: at the bottom of the page in the danger zone. Deletion cannot be undone.
Changes are saved automatically, a save button is not needed.
Tab: Parameters
Section titled “Tab: Parameters”This is where the connection settings are configured.
For template-based tools:
- The parameter fields are predefined by the template
- Enter the required values (e.g. server URL)
- Secret values (e.g. API keys, passwords) are stored as a connection: enter them once, after that they can no longer be viewed, only replaced or deleted
- An integration guide explains step by step what to enter
For manually created MCP servers:
- Server URL: the address of the MCP server (required field)
- Healthcheck URL: optional endpoint for availability checks
- Define parameters: add your own parameters (text, number, or yes/no). For each parameter you can decide how it is passed to the server (as an argument, HTTP header or query parameter).
Test connection:
Click on “Test connection” to check whether the connection works:
- Success (green checkmark): the connection works and the available capabilities have been detected. The response time is shown in milliseconds.
- Failed (red X): the connection could not be established. The error message gives hints for fixing it.
Important: The tool can only be activated after the connection has been tested successfully.
Tab: Connections
Section titled “Tab: Connections”Shows the connections of this tool, that is, the securely stored credentials. Here you can test and renew them and, for OAuth services, restart the sign-in with “Reconnect”. You find the same connections under Settings → Security in the “Credentials” section.
For skill integrations this is also where you define who may use the tool, via the connection’s access: whoever may co-use the connection can use the skill.
Tab: Sharing (MCP servers only)
Section titled “Tab: Sharing (MCP servers only)”Defines who can use this tool:
| Option | Description |
|---|---|
| Only me personally | Only you can use the tool (default) |
| Specific people or groups | You select which colleagues or teams get access |
| Entire tenant | All users in your workspace (administrators only) |
| Entire 9brains | Available system-wide for all users (super administrators only) |
Add people or groups:
- Select “Specific people or groups”
- Search for the name of the person or group
- Select the entry, it appears as a label
- To remove it, click the X next to the name
Tab: Capabilities (MCP servers only)
Section titled “Tab: Capabilities (MCP servers only)”Shown after a successful connection test. Lists all capabilities (tools) that the MCP server provides.
For each capability you can set the permission:
| Permission | Meaning |
|---|---|
| Always allowed | The AI may use this capability at any time |
| Ask first | The AI asks you in the chat before using the capability |
| Not allowed | The AI may not use this capability |
| Locked | Set by an administrator, cannot be changed |
Example: You want the AI to be able to read emails but always ask before sending. Then set “read email” to Always allowed and “send email” to Ask first.
Tab: Audit
Section titled “Tab: Audit”Shows when the AI called this tool, with search and filter options. You see your own calls; administrators see the calls of all workspace members.
Connecting Microsoft 365
Section titled “Connecting Microsoft 365”The Microsoft 365 integration gives the AI access to emails, calendars, OneDrive, Teams, contacts, tasks and other Microsoft services. Each user sets up their own personal connection.
Prerequisite: administrator authorization
Section titled “Prerequisite: administrator authorization”Before users can set up their integration, an administrator must grant the Microsoft authorization once. Details under Microsoft authorization for administrators. If the authorization has already been granted, this step can be skipped.
Set up the integration
Section titled “Set up the integration”- Go to Settings → System integrations
- Click on ”+ Tool” → “From template” → “Microsoft 365”
- A sign-in popup opens: sign in with your Microsoft 365 account and allow access
The connection is active immediately. The AI can now access your emails, calendars, files and other Microsoft services. Afterwards you find the connection under Settings → Security in the “Credentials” section, where you can test, reconnect or remove it. There you also see which Microsoft account it is connected with.
Troubleshooting Microsoft 365
Section titled “Troubleshooting Microsoft 365”If the AI can no longer access emails, calendars or other Microsoft services:
- Check your connection: Go to Settings → Security, open the “Microsoft 365” connection in the “Credentials” section and click “Test”. If no connection exists, set it up again (see above).
- Reconnect: If the test fails or the connection shows “Error”, click “Reconnect” and sign in with Microsoft again.
- Check administrator authorization: Only if the message “Administrator approval required” appears during sign-in does an administrator need to grant the Microsoft authorization.
Authentication with external services
Section titled “Authentication with external services”Depending on the integration, there are different sign-in methods:
Skill templates and connected skills: connections
Section titled “Skill templates and connected skills: connections”Credentials for skill templates and connected skills are stored as connections: you enter them once in a secure dialog, after which they are stored encrypted and can never be displayed by anyone again, only replaced or deleted. Testing, renewal and the activity log are available under Settings → Security in the “Credentials” section.
OAuth 2.0 (e.g. Microsoft 365, Google Workspace)
Section titled “OAuth 2.0 (e.g. Microsoft 365, Google Workspace)”OAuth enables a secure sign-in without you having to enter your password in 9brains: you sign in directly with the provider in a popup and allow access. The sign-in is renewed automatically as long as the connection exists. To revoke access, delete the connection under Settings → Security.
API key for MCP servers (e.g. Notion)
Section titled “API key for MCP servers (e.g. Notion)”Manually set up MCP servers use parameters in the tool editor:
- Create the API key in the respective service (the guide in the “Parameters” tab explains how)
- Enter the key in the corresponding field
- Test the connection
These keys are also stored as a connection: encrypted and no longer viewable after saving. Testing and renewal happen in the tool’s “Connections” tab or under Settings → Security.
Security & privacy
Section titled “Security & privacy”Credentials
Section titled “Credentials”- All credentials (API keys, OAuth tokens) are encrypted in the database
- Credentials are never shown in plain text in the interface
- OAuth tokens are renewed automatically as long as the connection is active
Skills and code execution
Section titled “Skills and code execution”- Skill code runs in a fully isolated sandbox environment
- The sandbox has no access to other user data or internal systems
- Each chat session has its own isolated working area
- After the session ends, the sandbox environment is destroyed
Connected skills (skills with external access)
Section titled “Connected skills (skills with external access)”- When a skill accesses external services (e.g. your CRM), the credentials from your connection are only inserted automatically when a network call is made
- The code in the sandbox never sees the credentials directly, and they are only ever sent to the service they were stored for
- Even the AI itself never knows your credentials: your input bypasses it and goes straight into encrypted storage