AI agents and tools need to reference your enterprise content to be impactful. As partners continue to add support for more connectors and integrations, faster than ever, AI providers are integrating with Box in increasingly different ways. This document provides an overview of common types of AI integrations, clarifies how they work, explains associated impacts on security, governance, and cost of using Box.
Important
These integrations are usually developed by our partners using Box public APIs. In such cases, Box does not always have the knowledge on how they were built or how they operate.
Common types of AI integrations
The table contains basic information on types of AI integrations. See more details in respective sections.
| AI Search Integration | AI Indexing Integration | Box MCP Server Integration | |
| Description | A third party AI agent searches the user’s Box data in real time | The user’s Box content is indexed, copied, stored, and updated inside the third party AI agent’s environment | A third party AI agent with an MCP client interacts with the Box MCP server using specific tools exposed by Box |
| Security |
|
|
|
| Cost | Low | High | Low |
AI Search Integration
A third party AI agent searches the user’s Box data in real time, calling Box APIs whenever the user makes a relevant query.
Security
User files and data stay inside Box unless explicitly returned in a query. Users can only access content they have permission to access in Box. Permissions are checked in real-time.
Cost
Typically lower costs because API calls only occur at the time of each relevant user query. User queries translate into API calls, which are typically billed against the API call limits and/or AI units allotted within their existing Box plan.
AI Indexing Integration
The entirety (or a significant portion) of the user’s Box content is indexed, copied, stored, and continuously updated inside the third party AI agent’s environment.
Security
User files and data are copied and stored in an external environment, which must be separately secured and governed by the third party. Many indexing connectors have checks in place to ensure users can only interact with the content they have permission to access in Box. However, those permission checks are managed by the third party in their environment and can be out of sync with Box. Important Review the third party documentation for the integration to ensure it meets your organization’s security requirements.
Cost
Typically higher costs because the provider makes bulk API calls to copy and store all of the user’s files and recurring API calls to keep that index updated. This can result in an enterprise unexpectedly exceeding the API limits allotted to their existing Box plan. User queries are then answered using the pre-built index of content that the third-party agent has copied and stored.
Box MCP Server Integration
A third party AI agent with an MCP client interacts with the Box MCP server using specific tools exposed by Box. The server calls Box APIs whenever the user makes a relevant query.
Security
User files and data stay inside Box unless explicitly returned in a query. Users can only access content they have permission to access in Box. Permissions are checked in real-time. This integration uses API calls to establish the connection between the third party MCP client and the Box MCP server, which are also typically billed against the API call limits allotted with their existing Box plan.
Cost
Typically lower costs because API calls only occur at the time of each relevant user query. User queries translate into API calls, which are typically billed against the API call limits and/or AI units allotted with their existing Box plan. Other API calls are used to establish the connection between the third party MCP client and the Box MCP server, which are also typically billed against the API call limits allotted with their existing Box plan. Existing AI integrations This list does not capture all AI integrations with Box and is subject to updates as new integrations are built and released.
| AI Search Integration | AI Indexing Integration | Box MCP Server Integration | Links | |
| Amazon Quick Suite | Yes | |||
| Anthropic Claude Connectors | Yes | |||
| Anthropic Claude Enterprise Search | Yes | |||
| Anthropic Claude Messages API | Yes | |||
| Fin by Intercom | Yes | |||
| GitHub MCP Registry | Yes | |||
| Glean MCP Directory | Yes | |||
| Glean Search | Yes | |||
| Google Gemini Enterprise Agent (Preview) | Yes | |||
| Google Gemini Enterprise Datasource Connector | Yes | |||
| Groq Responses API | Yes | |||
| IBM watsonx Orchestrate Agent | Yes | |||
| Microsoft Agent 365 (Preview) | Yes | |||
| Microsoft Azure API Center | Yes | |||
| Microsoft 365 Copilot Agent (Preview) | Yes | |||
| Microsoft Copilot Studio Tools | Yes | |||
| Microsoft Foundry Tools | Yes | |||
| Microsoft Copilot Connector | Yes | |||
| Microsoft Windows On-Device Agent Registry (ODR) | Yes | |||
| Mistral AI Le Chat | Yes | |||
| Notion AI | Yes | |||
| OpenAI ChatGPT App | Yes (“On demand”)* | Yes (“Sync”)* | ||
| OpenAI Company Knowledge | Yes | |||
| OpenAI Platform - AgentKit | Yes | |||
| Perplexity AI | Yes | |||
| ServiceNow - Agents | Yes | |||
| ServiceNow - Moveworks | Yes | |||
| Slack AI App | Yes |
*If you are a ChatGPT Pro, Business, or Enterprise user, both an on-demand connector and a sync connector are created when you connect to Box. You can disable sync during initial setup by opening Advanced settings, selecting On-demand, and clicking Save changes. You can also disable sync after the connector has been created by going to Settings → Apps → Box, clicking the three-dot menu, and selecting Disable sync.
If you are an admin on a ChatGPT Enterprise plan, you can also disable sync for your entire workspace. To do this, go to Account → Workspace settings → Apps and locate the Box connector. If you see an Enable sync button, sync is already disabled. If not, find the “Box sync configured” row, open the three-dot menu on the right, and select Delete to disable sync.
Alternatively, if you prefer not to use the built-in ChatGPT connector, you can manually configure an MCP-based Box connector that provides broader tool access without indexing content (instructions here). As an admin you can publish this MCP connector to your ChatGPT workspace so that employees within your workspace can use it without turning on dev mode.