OpenAI has introduced a new Developer mode for ChatGPT, giving users full access to the model context protocol (MCP) client across all tools. The feature, currently in beta, is available to Plus and Pro subscribers on the web.
According to OpenAI, the Developer mode enables both read and write capabilities for connectors, but also carries risks. “It’s powerful but dangerous and is intended for developers who understand how to safely configure and test connectors,” the company said in its guidance.
Users can enable the feature through Settings → Connectors → Advanced → Developer mode. Once active, developers can import MCPs by adding a remote MCP server in the Connectors tab. Supported protocols include SSE and streaming HTTP, with OAuth or no authentication options.
The feature allows developers to manage tools directly from connectors. They can toggle tools on or off, refresh connector data and use custom prompting to call specific tools in conversations. OpenAI cautions that careful tool selection is required to avoid ambiguity or misuse.
“Be explicit: ‘Use the Acme CRM connector’s update_record tool to…’,” the company advised, noting that developers should also specify sequencing of tool calls and state preferences when multiple connectors overlap.
Write actions triggered through Developer mode require user confirmation. OpenAI recommends closely inspecting JSON payloads before approval. “Incorrect write actions can inadvertently destroy, alter or share data,” the company warned.
The system respects the readOnlyHint annotation, treating tools without it as write actions. Users can choose to remember approval or denial for tools within a conversation, though OpenAI emphasised that such trust should be granted only to reliable applications.
Developer mode also provides visibility into tool input and output data, enabling debugging and verification of tool behaviour.
Your reaction
Discussion
What's your take on this story?
No comments yet — be the first to share your view.