ChatGPT Now Taps Company Data via MCP
Floriana/Getty Images
The Rise of Model Context Protocol
Anthropic's Model Context Protocol (MCP) is rapidly becoming the go to industry standard for smoothly integrating data with artificial intelligence systems. In an unusual display of collaboration among competitors, major tech companies including Google and OpenAI, as well as Microsoft, have embraced this open protocol. Its adoption in recent months stems from its capability to empower AI agents by connecting them to previously inaccessible data sources, a significant advantage for businesses dealing with siloed information.
OpenAI Integrates MCP into ChatGPT
Earlier this month, OpenAI incorporated MCP into ChatGPT, adding it to a suite of business oriented features. This integration allows users to connect their internal tools with OpenAI's Deep Research feature, enabling employees to access company specific data directly through the chatbot. OpenAI had initially adopted MCP in April, starting with its SDK. Currently, the functionalities are focused on search and document retrieval capabilities.
For further insights, you might also be interested in this comparison of ChatGPT's Deep Research against competitors like Gemini, Perplexity, and Grok AI.
(Disclosure: Ziff Davis, ZDNET's parent company, filed an April 2025 lawsuit against OpenAI, alleging it infringed Ziff Davis copyrights in training and operating its AI systems.)
How to Connect ChatGPT to Your MCP Server
In a detailed platform guide, OpenAI explains the setup process. Users need to build an MCP server compatible with Deep Research, utilizing search and fetch tools. Following this, a custom deep research connector must be created within ChatGPT. This connector requires thorough instructions to ensure the bot correctly interacts with your system. For users of ChatGPT Enterprise, Edu, and Team editions, there's an added benefit: they can publish this connector to their broader workspaces, where it can serve as a centralized knowledge source within Deep Research.
OpenAI also provides guidance for establishing authentication mechanisms and for testing the MCP server using the API Playground to ensure proper functionality before full deployment.
Navigating Risks and Best Practices
As with any emerging technology, implementing MCP comes with inherent risks and requires adherence to safety best practices. A key point to understand is that custom MCP servers are not vetted by OpenAI. Consequently, the company urges users to report any suspicious servers encountered to OpenAI via email at security AT openai DOT com. Such servers could potentially introduce prompt injections or malicious hidden directives that could manipulate ChatGPT's behavior.
For more on Large Language Models, consider reading about what Apple's controversial research paper reveals about LLMs.
OpenAI explicitly warns, "We recommend that you do not connect to a custom MCP server unless you know and trust the underlying application. For example, pick official servers hosted by the service providers themselves (e.g., we recommend connecting to the Stripe server hosted by Stripe themselves on mcp.stripe.com, instead of a custom Stripe MCP server hosted by a third party)."
It is crucial to remember that MCP servers grant OpenAI systems access to your company's data. Therefore, OpenAI strongly cautions users to meticulously review their tools and data sources for any sensitive information before enabling such connections.
