> ## Documentation Index
> Fetch the complete documentation index at: https://koreai.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

# Assistant Configurations

Assistant Configurations in the **Admin Hub** give administrators control over communication, security, pipeline behavior, and resource management across the AI deployment.

## Overview

| Configuration | Purpose |
| - | - |
| [Orchestration Settings](#orchestration-settings) | Configure the AI query processing pipeline |
| [Prebuilt Tools](#prebuilt-tools) | Configure the AI query processing pipeline |
| [Announcements](#announcements) | Publish banner messages to users |
| [Email Templates](#email-templates) | Customize invitation and notification emails for users and admins |
| [Guardrails](#guardrails) | Enforce PII detection and content restrictions |
| [MCP Server](#mcp-server) | Manage Model Context Protocol servers and agent tools |
| [Scheduler Settings](#scheduler-settings) | Manage automated agent execution permissions |
| [Language Settings](#language-settings) | Set a default language for the platform and new users |
| [Attachment Settings](#attachment-settings) | Control file attachment uploads and processing |
| [Projects Settings](#projects-settings) | Configure project features, model selection, and RAG capabilities |

## Orchestration Settings

Orchestration Settings is a pipeline management framework that lets administrators activate, configure, and manage individual components of the AI query processing workflow.

**Navigation**: Account Hub → Assistant Configurations → Orchestration Settings

![Orchestration Settings](https://mintcdn.com/koreai/tiqbbgo4pWG9ol09/ai-for-work/assist-configuration/images/orchestration-settings.png?fit=max\&auto=format\&n=tiqbbgo4pWG9ol09\&q=85\&s=759c9bc60a098a846a78629d404b6b37)

### Setup

1. Go to **Orchestration Settings** under Assist Configurations.
2. Review available pipeline components and their current status.
3. Configure components based on your organizational requirements.
4. Test the configuration with sample queries.
5. Monitor performance and adjust settings as needed.

### Pipeline Components

| Component | Function |
| - | - |
| **Guardrail Enforcement** | Applies safety and compliance checks to AI responses |
| **Small Talk Handling** | Enables casual, conversational responses for a natural user experience |
| **Intelligent Agent Routing** | Routes queries to the most appropriate agent based on intent |
| **Enterprise Knowledge Lookup** | Searches configured internal knowledge bases for accurate responses |
| **Fallback to AI Knowledge** | Uses LLM-derived knowledge when primary sources are unavailable |

### Guardrail Enforcement

Automatically applies security policies from your [Guardrails](#guardrails) configuration across all AI interactions.

![Guardrail Enforcement](https://mintcdn.com/koreai/Rd6io8e_F0P3SuW6/ai-for-work/assist-configuration/images/guardrails-enforcement.png?fit=max\&auto=format\&n=Rd6io8e_F0P3SuW6\&q=85\&s=5bed28050948bba5286f0c2baa86d8da)

* **Enabled**: Security policies enforce across all pipeline queries.
* **Disabled**: The pipeline skips enforcement.
* You cannot toggle this setting here; the [Guardrails](#guardrails) interface controls it exclusively.

### Small Talk Handling

Enables casual, contextual conversation for an improved user experience.

![Small Talk Handling](https://mintcdn.com/koreai/tiqbbgo4pWG9ol09/ai-for-work/assist-configuration/images/small-talk-handling.png?fit=max\&auto=format\&n=tiqbbgo4pWG9ol09\&q=85\&s=49e1765d3d0bdfa37f82f3a33e3fa92a)

* Toggle on or off based on organizational preferences.
* Uses professionally crafted, pre-configured prompts (not editable) to ensure a consistent brand voice.

### Intelligent Agent Routing

Analyzes incoming queries and directs them to the most appropriate AI agent.

![Intelligent Agent Routing](https://mintcdn.com/koreai/tiqbbgo4pWG9ol09/ai-for-work/assist-configuration/images/intelligent-agent.png?fit=max\&auto=format\&n=tiqbbgo4pWG9ol09\&q=85\&s=c839f84884cb6b31c98c26dac622d83d)

* Uses predefined routing algorithms optimized for accuracy and efficiency.
* The platform locks the routing logic to ensure consistent performance and scalability.

### Enterprise Knowledge Lookup

Integrates your organization's knowledge base into the AI query pipeline so responses draw from authoritative internal sources.

![Enterprise Knowledge Lookup](https://mintcdn.com/koreai/Rd6io8e_F0P3SuW6/ai-for-work/assist-configuration/images/enterprise-knowledge.png?fit=max\&auto=format\&n=Rd6io8e_F0P3SuW6\&q=85\&s=7fc9245b62a466e570d60345d13cdfe3)

| State | Behavior |
| - | - |
| **Configured and Enabled** | The pipeline queries the knowledge base automatically for all requests |
| **Configured but Disabled** | Knowledge base is accessible only through the compose bar's agent selector |
| **Not Configured** | The component remains inactive until you set up Enterprise Knowledge |

> [!NOTE]
>
> Enterprise Knowledge remains accessible through the compose bar's agent selector even when disabled in the query pipeline.

### Fallback to AI Knowledge

Provides responses from LLM knowledge when primary systems cannot respond adequately.

![Fallback to AI Knowledge](https://mintcdn.com/koreai/Rd6io8e_F0P3SuW6/ai-for-work/assist-configuration/images/fallback-ai.png?fit=max\&auto=format\&n=Rd6io8e_F0P3SuW6\&q=85\&s=9b05be2e0d9639880f3dde7660a346a6)

| Option | Description |
| - | - |
| **Toggle** | Enable or disable based on operational requirements |
| **LLM Model** | Select from available language models |
| **Web Search** | Optional integration; available for supported models only |
| **Custom Prompt** | Editable prompt for tailored AI response behavior |

***

## Prebuilt Tools

The Prebuilt Tools page provides a centralized view of platform tools available across all workspaces in your account.

**Navigation**: Admin Console > Assistant Configurations > Prebuilt Tools

The following tools appear on the page:

| Tool | Description |
| - | - |
| **AI Model** | Uses the AI model's general knowledge to generate answers; does not rely on your organization's internal data |
| **Web Search** | Allows end users to search the web |
| **Attachments** | Allows end users to attach files when asking queries |
| **Create** | Allows end users to generate PPTs, spreadsheets, docs, PDFs, and charts directly from their queries |

### AI Models

Administrators can add AI models from configured LLM connections and make them available to end users as selectable models.

**Add a model**

1. Click **Add Model** on the AI Model card.
2. In the popup, select a model from the list of all models configured in LLM Connections.
3. Provide a **Name** that end users see when selecting the model.
4. Provide a **Short Description** that explains the model's purpose to end users.
5. Click **Save**.

**Define publishing scope**

Set the visibility of each model:

| Scope | Description |
| - | - |
| **Everyone** | All users in the account can select the model |
| **Selected Users** | Only the specified users can select the model |
| **Selected User Groups** | Only members of the specified user groups can select the model |

***

## Enterprise Search

Enterprise Search lets administrators enable a dedicated search experience for end users.

**Navigation**: Admin Console > Assistant Configurations > Enterprise Search

Turn on the **Enable** toggle to make the **Search** option available to end users in the application.

When you enable Enterprise Search, select one enterprise knowledge source from the configured sources. The platform uses the selected source for all search results that end users see. You can change the selected source at any time.

***

## Announcements

The Announcements feature lets administrators create, manage, and publish organizational messages as banners at the top of the user interface.

**Navigation**: Account Hub → Assistant Configurations → Announcements

![Announcements](https://mintcdn.com/koreai/Rd6io8e_F0P3SuW6/ai-for-work/assist-configuration/images/announcement_1.png?fit=max\&auto=format\&n=Rd6io8e_F0P3SuW6\&q=85\&s=9fdec072c8b922466086617c16b209f5)

### Create an Announcement

1. Click **Create Announcement**.
2. Complete the required fields:

   | Field | Details |
   | - | - |
   | **Name** | Up to 80 characters |
   | **Description** | Up to 800 characters; supports bold, italic, underline, hyperlinks, colors, bullets, and numbered lists |
   | **Announcement Type** | Banner (currently the only option) |
   | **Publish To** | **Everyone in Account**, **Selected User Groups**, or **Agent Specific** |
3. If you select **Selected User Groups**, the platform displays all available user groups. Select one or more groups from the list; the interface displays the number of selected groups. The published announcement is visible to every user who belongs to the selected groups.

### Actions

| Action | Description |
| - | - |
| **Preview** | View the banner before publishing |
| **Publish** | Make the announcement visible (requires Name and Description) |
| **Unpublish** | Remove a published announcement from view |
| **Delete** | Permanently remove any announcement, published or not |

### Manage Published Announcements

Editing a published announcement prompts you to either **Discard changes** or **Publish changes**.

> [!NOTE]
>
> You must publish again to push changes to users. The system prompts you to publish or discard modifications to keep updates intentional and controlled.

![Published announcement management](https://mintcdn.com/koreai/Rd6io8e_F0P3SuW6/ai-for-work/assist-configuration/images/announcement_3.png?fit=max\&auto=format\&n=Rd6io8e_F0P3SuW6\&q=85\&s=6489ac8365390c6bf791c3cd92b3ed33)

### End-User Experience

* Announcements appear as banners at the top of the screen.
* Multiple announcements rotate every **10 seconds**.
* Each banner includes a **Dismiss** button:
  * Dismissing hides the banner for the current session.
  * The banner reappears after a page refresh or after one hour.

***

## Email Templates

Email Templates let you customize the emails that the platform sends when you add users to the account, a workspace, or specific applications, and when someone publishes an agent or flow to a user.

**Navigation**: Account Hub → Assistant Configurations → Email Templates

### Template Categories

#### Admin Templates

| Template | Sent When |
| - | - |
| **Admin Invitation** | You add a user as an Admin |
| **Custom Admin Invitation** | You assign a custom admin role to a user |
| **Workspace Invitation** | You invite a user to a specific workspace |

#### Application Templates

Each provisioned application has a **Member Invitation** template that the platform sends when you add a user as a member. Each application maintains its own separate template.

Each application also has an agent publish template. The platform sends this email, along with an in-app notification, when someone publishes an agent or flow to a user and keeps **Notify newly published users** selected in the publish dialog. The email includes the agent name, the agent description, whether the agent is enabled or disabled, and a **View agent** button. Turn off the template to stop these emails.

### Customize a Template

1. Use the toggle to **enable or disable** the template.
2. Click the template to open the editor.
3. Edit the **subject line** and **email body**.
4. **Preview** the email to verify appearance.
5. Add links to relevant resources (such as the Admin Console) for easy access.

### Template Management

* The platform organizes templates by application; all provisioned applications appear in the list.
* The platform provides default subject and body text, which you can modify.
* Changes apply to all future emails of that type.
* Admins and Custom Admins can review and adjust templates before the system sends emails.

***

## Guardrails

Guardrails is a multi-layered security and compliance framework that automatically scans AI inputs and outputs to protect sensitive data and enforce content policies.

**Navigation**: Account Hub → Assistant Configurations → Guardrails

MCP servers can also run these guardrails before any tool executes. For details, see [Tool guardrails](#tools).

![Guardrails overview](https://mintcdn.com/koreai/tiqbbgo4pWG9ol09/ai-for-work/assist-configuration/images/guardrails-implementation.png?fit=max\&auto=format\&n=tiqbbgo4pWG9ol09\&q=85\&s=514487d83a2fea19ec98ae073675cec5)

### Setup

1. Go to **Guardrails** in Account Hub.

   ![Guardrails in Account Hub](https://mintcdn.com/koreai/Rd6io8e_F0P3SuW6/ai-for-work/assist-configuration/images/guardrails-account.png?fit=max\&auto=format\&n=Rd6io8e_F0P3SuW6\&q=85\&s=f2e23c9ba734e128f211fb8d12b38d24)

2. Review pre-configured PII detection rules.

3. Set up content restrictions for your organization.

4. Validate configurations using the integrated testing suite.

### Interface Layout

The Guardrails interface has three sections: **PII Settings**, **Ban Topics**, and **Testing Suite**.

### PII Settings

Pre-configured rules detect common PII types:

| Rule | Detects |
| - | - |
| **Email Address** | Email patterns across various formats |
| **Universal Phone Numbers** | Domestic and international phone formats |
| **Social Security Numbers** | SSN patterns with format validation |
| **Credit Card Numbers** | Major credit card number formats |

![PII Settings](https://mintcdn.com/koreai/Rd6io8e_F0P3SuW6/ai-for-work/assist-configuration/images/PII-Settings.png?fit=max\&auto=format\&n=Rd6io8e_F0P3SuW6\&q=85\&s=4788e7be7299faa52ea8c57af6eb77bc)

#### Add or Update a Rule

Create custom detection patterns using regular expressions:

| Field | Description |
| - | - |
| **Name** | Unique identifier for the rule |
| **Description** | What the rule detects |
| **Regex** | Pattern used to match sensitive data |
| **Action** | How redacted data appears to non-authorized users |

**Action types:**

| Action | Behavior |
| - | - |
| **Redaction** | Removes sensitive data from text |
| **Static Replacement** | Substitutes with predefined safe text |
| **Masking** | Replaces with placeholder characters (e.g., `****` or `XXX`) |

![Action configuration](https://mintcdn.com/koreai/Rd6io8e_F0P3SuW6/ai-for-work/assist-configuration/images/action.png?fit=max\&auto=format\&n=Rd6io8e_F0P3SuW6\&q=85\&s=84bc140be0052e101f2b10893c0feeb7)

### Ban Topics

Ban Topics prevents AI models from engaging with prohibited subjects by dynamically modifying input prompts in real time, ensuring consistent compliance across all AI interactions.

| Feature | Description |
| - | - |
| **Template System** | Pre-defined restriction sets based on common industry standards |
| **Custom Prompts** | Create detailed, context-specific content restrictions |
| **Enable/Disable** | Toggle built-in or custom restrictions individually as needed |

### Testing Suite

Validate guardrail configurations in an isolated environment before deploying them:

* Run full or targeted validation of PII rules and banned topics.
* Use manual, AI-driven, or batch test inputs, including boundary condition testing.
* Analyze results by rule, impact status, processing duration, and output.
* Preview data sanitization visually.

![Testing framework](https://mintcdn.com/koreai/tiqbbgo4pWG9ol09/ai-for-work/assist-configuration/images/testing-framework.png?fit=max\&auto=format\&n=tiqbbgo4pWG9ol09\&q=85\&s=fec106e3b9d4b0941188d7b9a17254a5)

***

## MCP Server

The MCP Server feature lets administrators create and manage Model Context Protocol (MCP) servers that expose agents, prebuilt tools, and enterprise knowledge sources as tools. Administrators control which tools each server provides, how clients authenticate through OAuth 2.1, and which users can access the server. MCP-compatible clients connect to a server through its unique URL to run tasks and orchestrate workflows.

**Navigation**: Admin Console > Assistant Configurations > MCP Server

![MCP Server](https://mintcdn.com/koreai/is1XQsJMolvIATzd/ai-for-work/assist-configuration/images/mcp-servers-admin.png?fit=max\&auto=format\&n=is1XQsJMolvIATzd\&q=85\&s=638df05a9129a7c5ea5246ec3d17cac5)

### Server List

The MCP Server page lists every server in the account. Each server entry displays the creator, the creation date, the number of configured tools, and the authentication type. Use the **Search** box to find a server by name. Select **Create** to start the server creation wizard.

The page includes a pre-provisioned **Default MCP Server** that the platform marks with a **Default** label.

### Default MCP Server

Every account includes a **Default MCP Server** that requires no setup and is ready on first access. The Default MCP Server exposes the account's prebuilt agents and enterprise knowledge sources as MCP tools.

The platform provides the server name, "Default MCP Server", and the description, "Access your organization's tools, agents, and knowledge sources directly from your AI assistant." You cannot edit the name or description, and you cannot delete the Default MCP Server. You can edit its tools, authentication, and access limits. For details, see [Manage a server](#manage-a-server).

### Create a Server

Select **Create** to open the server creation wizard. The wizard guides you through four steps: **Details**, **Tools**, **Auth settings**, and **Access limits**.

#### Details

Enter a **Server Name** and a **Description**. Select **Continue** to create the server and generate its unique **Remote MCP URL**. The wizard displays this URL on the remaining steps, where you can copy it.

#### Tools

Configure the tools that the server exposes and whether guardrails run before tool execution.

**Tool guardrails**

The **Enable guardrail** toggle at the top of the Tools section controls whether the account guardrails run before any tool on the server executes.

| Toggle state | Behavior |
| - | - |
| **On** | The guardrails that administrators configure in [Guardrails](#guardrails) run before every tool execution, across all tools on the server. |
| **Off** | Tools run without the guardrail step. |

> [!NOTE]
>
> The toggle stays unavailable until you enable at least one guardrail for the account. Hover over the unavailable toggle to view the message "Guardrails disabled for the account."

**Add tools**

Select **Add tool** to open the tool picker. The picker has two tabs, **Prebuilt** and **Custom**, and opens on the **Prebuilt** tab. You can add tools from multiple systems and agents in a single flow.

To add prebuilt tools:

1. On the **Prebuilt** tab, select a system, such as `M365 Email`, `Jira`, `Zendesk`, or `Slack`, to view its tools.
2. Select one or more tools.
3. Select **Done**.
4. To add tools from another system, repeat these steps.

To add custom tools:

1. Select the **Custom** tab. The tab lists every agent in the account. The first five entries are the Enterprise Knowledge agents, followed by all other agents in alphabetical order.
2. Use the search box to find an agent.
3. Select one or more agents.
4. Select **Done**.

The Tools section lists every tool you add and shows the tool count split into **Prebuilt** and **Custom**. The platform pre-populates each custom tool with the agent's purpose as its tool description.

**Manage custom tools**

Hover over a custom tool to view the available actions:

| Action | Result |
| - | - |
| **Edit description** | Edit the tool description. The platform preserves edited descriptions within this server's configuration. |
| **Remove** | Remove the tool from the server. To add the tool again, select **Add tool** and select it from the **Custom** tab. |

**Manage prebuilt tools**

The Tools section groups prebuilt tools by system.

* To remove a system and all its tools in one action, hover over the system and select **Remove**. The button shows the number of tools added from that system.
* To remove individual tools, select the system to open a side panel that lists its selected tools. Hover over a tool and select **Remove**.
* To add more tools from the same system, select **Add more tools from \<system name>** in the side panel. The tool picker opens on that system and lists all its available tools.

#### Auth settings

Configure how MCP clients register with the server and authenticate users. AI for Work MCP servers support OAuth 2.1 authorization, as defined in the [MCP authorization specification](https://modelcontextprotocol.io/specification/2025-11-25/basic/authorization). Every MCP client must authenticate through OAuth before it can list or run the server's tools.

**How OAuth works with the MCP server**

1. The MCP client sends a request to the server's **Remote MCP URL**. The server responds that authorization is required and points the client to the AI for Work authorization server.
2. The client identifies itself through one of the registration methods that you configure on this step: **Client ID Metadata Document (CIMD)** or **Static OAuth**.
3. The client opens a browser window, and the user signs in to AI for Work and approves access for the client.
4. AI for Work issues an access token to the client. The client sends this token in the `Authorization: Bearer` header with every request to the server.
5. The server runs tools on behalf of the signed-in user, within that user's existing permissions and the server's [Access limits](#access-limits).

The flow follows the OAuth 2.1 authorization code flow with PKCE. For protocol details, see [Authorization](https://modelcontextprotocol.io/specification/2025-11-25/basic/authorization) in the MCP specification.

**Registration methods**

MCP clients differ in how they register with an OAuth server. Choose the method that your MCP client supports:

| Method | How the client registers | Use when |
| - | - | - |
| **Client ID Metadata Document (CIMD)** | The client identifies itself with an HTTPS URL that hosts its metadata document. You don't create a client in AI for Work. | The MCP client supports CIMD and connects without pre-registered credentials. |
| **Static OAuth** | You create a client in AI for Work, and then enter its **Client ID** and **Client Secret** in the MCP client. | The MCP client requires a pre-registered client ID and secret, such as Google Gemini. |

> [!NOTE]
>
> The MCP authorization specification deprecates Dynamic Client Registration (DCR) and recommends CIMD for new implementations. AI for Work uses CIMD for clients that connect without a pre-created client. For details, see [Client Registration](https://modelcontextprotocol.io/specification/2026-07-28/basic/authorization/client-registration) in the MCP specification. To confirm which method a client supports, see the client's documentation.

**Client ID Metadata Document (CIMD)**

With CIMD, the MCP client uses an HTTPS URL as its client ID. The URL points to a JSON metadata document that the client publishes, which lists the client name and its redirect URLs. When the client connects, AI for Work fetches the document, verifies that the client's redirect URL matches an allowed domain, and shows the client name on the consent screen. For the document format, see [OAuth Client ID Metadata Document](https://datatracker.ietf.org/doc/html/draft-ietf-oauth-client-id-metadata-document-00).

To configure CIMD:

1. Turn on the **CIMD** toggle. The **Allowed Domains** button appears.
2. Select **Allowed Domains** to view the pre-configured domains. All pre-configured domains are enabled by default.
3. Enable or disable each domain.
4. Optional: add a custom domain to allow a client that the list does not include. For the format, see **Domain pattern format** in this section.

You can delete custom domains only. To block a pre-configured domain, disable it.

**Pre-configured allowed domains**

The platform allows the following domain patterns by default. Each pattern matches the redirect URLs of a supported MCP client.

| Type | Domain pattern | MCP client |
| - | - | - |
| Local | `http://127.0.0.1:*/**` | Desktop apps and CLI tools that receive the OAuth callback on the user's device |
| Local | `http://localhost:*/**` | Desktop apps and CLI tools that receive the OAuth callback on the user's device |
| Local | `http://[::1]:*/**` | Desktop apps and CLI tools that receive the OAuth callback on the user's device (IPv6) |
| HTTPS | `https://chatgpt.com/**` | ChatGPT |
| HTTPS | `https://claude.ai/**` | Claude |
| HTTPS | `https://claude.com/**` | Claude |
| HTTPS | `https://vscode.dev/**` | Visual Studio Code |
| HTTPS | `https://vertexaisearch.cloud.google.com/**` | Google Vertex AI Search |
| HTTPS | `https://*.perplexity.ai/**` | Perplexity |
| HTTPS | `https://*.perplexity.com/**` | Perplexity |
| HTTPS | `https://callback.mistral.ai/**` | Mistral AI |
| HTTPS | `https://api.devin.ai/**` | Devin |
| HTTPS | `https://app.kiro.dev/agent/mcp/callback` | Kiro |
| HTTPS | `https://lovable.dev/**` | Lovable |
| HTTPS | `https://manus.im/api/webhook/mcp/callback` | Manus |
| HTTPS | `https://app.writer.com/**` | Writer |
| HTTPS | `https://api.stilla.ai/**` | Stilla |
| HTTPS | `https://global.consent.azure-apim.net/**` | Microsoft Power Platform and Copilot Studio |
| HTTPS | `https://token.botframework.com/**` | Microsoft Bot Framework |
| HTTPS | `https://us-east-1.quicksight.aws.amazon.com/**` | Amazon QuickSight |
| HTTPS | `https://*.azuredatabricks.net/**` | Azure Databricks |
| HTTPS | `https://*.databricks.com/**` | Databricks |
| HTTPS | `https://*.snowflake.app/**` | Snowflake |
| HTTPS | `https://*.apps.dynatrace.com/**` | Dynatrace |
| HTTPS | `https://*.resolve.ai/**` | Resolve AI |
| HTTPS | `https://*.natoma.app/**` | Natoma |
| HTTPS | `https://b.app.lumosidentity.com/b/oauth/authorize` | Lumos |
| HTTPS | `https://b.app.lumosidentity.com/b/oauth/token` | Lumos |
| HTTPS | `https://mcp.docker.com/**` | Docker |
| HTTPS | `https://oauth.pstmn.io/**` | Postman |
| HTTPS | `https://integrations.zoom.us/**` | Zoom |
| HTTPS | `https://oauth2.slack.com/external/auth/**` | Slack |
| HTTPS | `https://app.asana.com/-/integrations/oauth/callback` | Asana |
| HTTPS | `https://www.canva.com/**` | Canva |
| HTTPS | `https://figma-gov.com/**` | Figma for Government |
| HTTPS | `https://api.gamma.app/mcp/atlassian_mcp/callback` | Gamma |
| HTTPS | `https://coda.io/**` | Coda |
| HTTPS | `https://docs.superhuman.com/**` | Superhuman |
| HTTPS | `https://*.kore.ai/**` | Kore.ai |
| Custom scheme | `cursor://cursor.mcp` | Cursor |

> [!NOTE]
>
> The local patterns let MCP clients that run on the user's device, such as CLI tools and desktop apps, complete the OAuth flow. Keep them enabled to support these clients, or disable them to restrict access to hosted clients only.

**Domain pattern format**

Add a custom domain as an exact URL or as a pattern that covers several subdomains, ports, or paths.

| Element | Meaning | Example |
| - | - | - |
| Protocol | Required in every entry. Use `https://`, `http://` for local addresses, or the client's custom scheme. | `https://`, `cursor://` |
| `*` in the host | Matches any single subdomain. | `https://*.example.com/**` |
| `:*` after the host | Matches any port. | `http://localhost:*/**` |
| `/**` after the host | Matches any path under the domain. | `https://app.example.com/**` |
| Exact path | Matches only that path. | `https://app.example.com/oauth/callback` |

For a comparable reference on domain allowlists for MCP OAuth clients, see [Available Atlassian MCP server domains](https://support.atlassian.com/security-and-access-policies/docs/available-atlassian-mcp-server-domains/).

**Static OAuth**

Static OAuth lets you register specific clients and issue each client its own credentials.

To add a static client:

1. Under **Static OAuth**, select **Add Client**.
2. Enter a **Client Name**.
3. Enter one or more **Redirect URL(s)**. To remove a redirect URL, delete it before you save the client.
4. Select **Generate Client ID and Secret**. The button stays unavailable until you enter the client name and a redirect URL. The platform generates a **Client ID** and a **Client Secret**.
5. Select **Done**. The client appears under **Static OAuth**.

To connect an MCP client that requires a static client, such as Google Gemini:

1. In the MCP client's documentation or settings, find the OAuth redirect (callback) URL that the client uses.
2. In AI for Work, add a static client and enter that URL in **Redirect URL(s)**. For steps, see the previous procedure.
3. Copy the **Client ID** and **Client Secret**, and the **Remote MCP URL** from the wizard.
4. In the MCP client, add the server: enter the **Remote MCP URL**, and then enter the **Client ID** and **Client Secret** in the client's OAuth configuration. For client-specific steps, see [Connect an MCP client](#connect-an-mcp-client).
5. Connect to the server from the MCP client and sign in when the client prompts you.

To manage an existing static client:

| Action | Result |
| - | - |
| **Edit** | Update the client name or redirect URLs. |
| **Regenerate** | Generate a new **Client Secret** for the client. |
| **Delete** | Remove the client from the server. |

#### Access limits

Set which users in the account can access the server:

| Option | Access |
| - | - |
| **Everyone** | All users in the account can access the server. This is the default. |
| **Limited users** | Only the selected users or user groups can access the server. Use the directory picker to search for and select users and groups. |
| **None** | No users can access the server, which makes the server inaccessible. |

### Connect an MCP client

After you create the server, add it to an MCP client so that users can run its tools from that client.

**Before you begin**

* Copy the server's **Remote MCP URL** from the wizard.
* Confirm that the user has access to the server under [Access limits](#access-limits).
* For a client that uses CIMD, confirm that the client's domain is enabled under **Allowed Domains**.
* For a client that uses Static OAuth, create a static client and copy its **Client ID** and **Client Secret**. Register the client's redirect URL listed in the following sections.

In the following examples, replace `<Remote MCP URL>` with your server's URL and `ai-for-work` with a name of your choice. Client menus and commands change over time; for the latest steps, see each client's documentation.

> [tab] Claude
>
> Custom connectors are available on the Claude web and desktop apps for Pro, Max, Team, and Enterprise plans.
>
> 1. In Claude, go to **Settings** > **Connectors**.
> 2. Select **Add custom connector**.
> 3. Enter a name and the **Remote MCP URL**.
> 4. For Static OAuth only: select **Advanced settings**, and then enter the **Client ID** and **Client Secret**.
> 5. Select **Add**.
> 6. Select **Connect**, and then sign in to AI for Work when prompted.
>
> On Team and Enterprise plans, only an Owner can add the connector. Members select **Connect** on the custom connector to authenticate.
>
> For details, see [Get started with custom connectors using remote MCP](https://support.claude.com/en/articles/11175166-getting-started-with-custom-connectors-using-remote-mcp).

> [tab] Claude Code
>
> 1. Add the server:
>
>    ```bash theme={null}
>    claude mcp add --transport http ai-for-work <Remote MCP URL>
>    ```
>
> 2. Start Claude Code, run `/mcp`, select the server, and complete the sign-in in the browser.
>
> For Static OAuth, register `http://localhost:8080/callback` as the redirect URL of the static client, and then add the server with the client credentials. The command prompts for the secret.
>
> ```bash theme={null}
> claude mcp add --transport http \
>   --client-id <Client ID> --client-secret --callback-port 8080 \
>   ai-for-work <Remote MCP URL>
> ```
>
> For details, see [Connect Claude Code to tools via MCP](https://code.claude.com/docs/en/mcp).

> [tab] ChatGPT
>
> Custom MCP apps are available on ChatGPT web for Business, Enterprise, and Edu plans, with limited support on Pro. An admin must enable developer mode first.
>
> 1. Turn on developer mode under **Settings** > **Apps** > **Advanced settings**.
> 2. Go to **Settings** > **Apps**, and then select **Create**.
> 3. Enter a name and the **Remote MCP URL**.
> 4. Select **OAuth** as the authentication method.
> 5. Select **Scan Tools**, and then sign in to AI for Work when prompted.
> 6. Select **Create**.
>
> For details, see [Developer mode and MCP apps in ChatGPT](https://help.openai.com/en/articles/12584461-developer-mode-apps-and-full-mcp-connectors-in-chatgpt-beta).

> [tab] Cursor
>
> Add the server to `.cursor/mcp.json` in the project root, or to `~/.cursor/mcp.json` to make it available in all projects:
>
> ```json theme={null}
> {
>   "mcpServers": {
>     "ai-for-work": {
>       "url": "<Remote MCP URL>"
>     }
>   }
> }
> ```
>
> Cursor prompts you to sign in to AI for Work when you first use the server.
>
> For Static OAuth, register the following redirect URLs on the static client. Cursor uses the same fixed redirect URLs for every MCP server.
>
> | Redirect URL | Cursor surface |
> | - | - |
> | `http://localhost:8787/callback` | Desktop app |
> | `https://www.cursor.com/agents/mcp/oauth/callback` | Web and Cursor Agents |
>
> Set the `AIW_CLIENT_ID` and `AIW_CLIENT_SECRET` environment variables to the client credentials, and then add an `auth` block. When the `auth` block contains a client ID, the Cursor desktop app sends only the desktop redirect URL.
>
> ```json theme={null}
> {
>   "mcpServers": {
>     "ai-for-work": {
>       "url": "<Remote MCP URL>",
>       "auth": {
>         "CLIENT_ID": "${env:AIW_CLIENT_ID}",
>         "CLIENT_SECRET": "${env:AIW_CLIENT_SECRET}"
>       }
>     }
>   }
> }
> ```
>
> > [!NOTE]
> >
> > If sign-in fails with a redirect URL error, open the **Output** panel in Cursor, select **MCP Logs**, and check the `redirect_uri` that Cursor sent. Some Cursor desktop versions still send the earlier custom-scheme URL, `cursor://anysphere.cursor-mcp/oauth/callback`. Add the URL from the log to the static client. The desktop callback also fails if another application on the device already uses port 8787.
>
> For details, see [Model Context Protocol (MCP)](https://cursor.com/docs/mcp) in the Cursor documentation.

> [tab] VS Code
>
> 1. Open the Command Palette and run **MCP: Add Server**, or add the server to `.vscode/mcp.json`:
>
>    ```json theme={null}
>    {
>      "servers": {
>        "ai-for-work": {
>          "type": "http",
>          "url": "<Remote MCP URL>"
>        }
>      }
>    }
>    ```
>
> 2. Start the server and sign in to AI for Work when VS Code prompts you.
>
> For details, see [Add and manage MCP servers in VS Code](https://code.visualstudio.com/docs/copilot/customization/mcp-servers).

> [tab] Codex
>
> 1. Add the server:
>
>    ```bash theme={null}
>    codex mcp add ai-for-work --url <Remote MCP URL>
>    ```
>
> 2. Sign in:
>
>    ```bash theme={null}
>    codex mcp login ai-for-work
>    ```
>
> For details, see [Model Context Protocol](https://developers.openai.com/codex/mcp) in the Codex documentation.

> [tab] Gemini CLI
>
> Gemini CLI uses Static OAuth. Register `http://localhost:7777/oauth/callback` as the redirect URL of the static client, and then add the server to `~/.gemini/settings.json`:
>
> ```json theme={null}
> {
>   "mcpServers": {
>     "ai-for-work": {
>       "httpUrl": "<Remote MCP URL>",
>       "oauth": {
>         "enabled": true,
>         "clientId": "<Client ID>",
>         "clientSecret": "<Client Secret>",
>         "redirectUri": "http://localhost:7777/oauth/callback"
>       }
>     }
>   }
> }
> ```
>
> Run `/mcp auth ai-for-work` in Gemini CLI, and then sign in to AI for Work in the browser.
>
> For details, see [MCP servers with the Gemini CLI](https://geminicli.com/docs/tools/mcp-server/).

> [!TIP]
>
> For other MCP clients, enter the **Remote MCP URL** in the client's remote MCP server settings, and use OAuth as the authentication method. If the client requires a client ID and secret, create a static client first.

### Manage a server

You can edit any server after you create it, including the **Default MCP Server**. Open a server to modify its tools, authentication method, or access limits, and the details of servers that you create. You can delete any server except the **Default MCP Server**.

***

## Scheduler Settings

Scheduler Settings lets administrators configure account-wide controls for automated agent execution, including availability, usage limits, and workspace permissions.

**Navigation**: Account Hub → Assistant Configurations → Scheduler Settings

### Enable Scheduler

Control whether users can access scheduling functionality for agents.

| State | Effect |
| - | - |
| **On** | Users can create and manage schedulers for their agents |
| **Off** | The platform disables scheduling for all users |

### Scheduling Limit per User

Set the maximum number of active schedulers each user can create simultaneously:

* No Limit
* 5 schedulers
* 10 schedulers
* 20 schedulers
* 30 schedulers

Select the limit that aligns with your organization's automation needs and resource capacity.

### Workspace Owner Permissions

When enabled, workspace owners can:

* Add schedulers during the agent creation workflow.
* Publish agents with pre-configured schedules.
* Set schedulers to run automatically for all end users.

End users receive agents with active schedulers but retain control to disable any scheduler they don't need.

For agent-level scheduler configuration, see [Schedule Trigger](https://koreai.mintlify.site/ai-for-work/custom-agents/agent-management.md#schedule-trigger).

***

## Language Settings

The platform supports the following languages:

| Language | Code |
| - | - |
| English (US) | `en` |
| Arabic | `ar` |
| German | `de` |
| Spanish | `es` |
| French | `fr` |
| Japanese | `ja` |
| Thai | `th` |

Administrators can set a default language for the account. Users can override this selection from **Settings** > **Language** in the end-user application.

***

## Attachment Settings

Attachment Settings let administrators control whether end users can upload file attachments in the compose bar and configure how the platform processes those attachments.

**Navigation**: Account Hub → Assistant Configurations → Attachment Settings

### Enable Uploading Attachments in Compose Bar

Use the **Enable Uploading Attachments in Compose Bar** toggle to control whether end users can upload files when asking queries.

| State | Effect |
| - | - |
| **On** | End users see the attachment option in the compose bar and can upload files alongside their queries |
| **Off** | The platform hides the attachment option from the compose bar for all end users |

> [!NOTE]
>
> You can enable attachment uploads only if the platform has at least one configured LLM model. If no LLM model is available, the toggle remains disabled and the platform displays a banner prompting you to configure an LLM model first.

### LLM Model Selection

When you enable attachment uploads, the platform displays an **LLM model selection** dropdown below the toggle.

* The platform selects the **default LLM model** automatically.
* You can change the LLM model used for processing attachments at any time.
* Changes to the LLM model take effect immediately for all subsequent queries.

### Allow Attachments Larger Than Model Context

A separate toggle controls whether end users can upload files that exceed the selected model's context size.

| State | Behavior |
| - | - |
| **Off** | The platform restricts uploaded files to the selected model's context size limit |
| **On** | The platform accepts files larger than the model's context size and processes them using RAG (Retrieval-Augmented Generation) |

When enabled, select one of the following RAG processing options:

| Option | Description |
| - | - |
| **Platform Built-in RAG** | Uses the platform's built-in RAG capability to process and retrieve content from large attachments. Displays a **Select Embedding Model** dropdown; the platform selects the Kore embedding model by default. Only embedding-capable models appear in the list. |
| **LLM-Provided RAG** | Uses the selected LLM's native RAG or file-handling capabilities. Available only if the chosen LLM model supports the required APIs. |

> [!NOTE]
>
> Switching between RAG options or changing the embedding model removes all currently uploaded files across the account. The platform prompts you to confirm before applying the change.

***

## Personalisation and Memory

Personalisation and Memory lets administrators configure the assistant's ability to generate user memory and deliver personalized responses.

**Navigation**: Admin Console > Assistant Configurations > Personalisation and Memory

![Personalise](https://mintcdn.com/koreai/_HfvtzXTpqe0jmK_/ai-for-work/assist-configuration/images/personalise.png?fit=max\&auto=format\&n=_HfvtzXTpqe0jmK_\&q=85\&s=c52d71f85ae5402453727cb5a4906421)

The page provides three account-wide toggles:

| Setting | Description |
| - | - |
| **Reference Past Chats?** | The assistant references the user's previous conversations for context |
| **Capture Memory?** | The assistant remembers information the user shares across conversations |
| **Mine Data from Business Systems?** | The assistant fetches details from connected business applications |

When an administrator turns on **Capture Memory**, all existing end users see a one-time pop-up notification in the AI for Work app confirming that memory is active. Any new user who receives access while memory is active also sees this notification on first login.

> [!NOTE]
>
> Turning off Capture Memory stops new memory generation but does not delete existing memories. Users can manage their stored memories from the Personalisation and Memory widget on their profile page.

***

## Projects Settings

Projects Settings lets administrators configure project features, model selection, and RAG capabilities for project-based queries and file processing.

**Navigation**: Account Hub → Assistant Configurations → Projects Settings

### Enable Projects

Use the **Enable Projects** toggle to control whether users can create and manage projects.

| State | Effect |
| - | - |
| **On** | Users can create projects, upload files, and use project content as a source for queries |
| **Off** | The platform hides the Projects feature from all users |

### Model Selection

When you enable Projects, the platform displays a **Model Selection** dropdown.

* The platform selects the default configured LLM model automatically.
* You can change the model used for processing project-based queries at any time.
* Models that support RAG display a **Supports RAG** label next to the model name.

### RAG Configuration

RAG (Retrieval-Augmented Generation) allows the platform to query large attachments that exceed context limits and to execute queries on project content.

Select one of the following RAG processing options:

| Option | Description |
| - | - |
| **Platform Built-in RAG** | Uses the platform's built-in RAG capability to process and retrieve content from large attachments within projects |
| **LLM-Provided RAG** | Uses the selected LLM's native RAG or file-handling capabilities to process large attachments. Available only if the chosen model supports the required APIs |

> [!NOTE]
>
> Switching between RAG options affects file processing across all projects in the account. The platform prompts you to confirm before applying the change.

***


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.