OpenAI has rolled out a series of new features for its Responses API, targeting developers and businesses building AI-powered applications with GPT-4o and the o-model family.
The Responses API serves as OpenAI's main interface for building "agentic" applications—systems that can handle tasks autonomously by calling tools, fetching information, or interacting with external services.
A major part of this update is support for remote MCP servers. The Model Context Protocol (MCP) is an open standard introduced by Anthropic that manages access to external tools and data sources, and it's being adopted by more AI companies. With the new update, developers can now connect any MCP-compatible server, including services from Stripe, Twilio, Shopify, PayPal, or Intercom.
New tools built into the Responses API
Several new tools are now integrated directly into the Responses API. This includes the gpt-image-1 image model, which powers image generation for both ChatGPT and Sora. The Code Interpreter from ChatGPT is now included as well, enabling complex calculations, data analysis, and image processing. File search has also been upgraded, with support for attribute-based filters and searching across multiple vector stores.
For agent developers, OpenAI has added new capabilities focused on reliability, privacy, and usability. The new background mode lets long-running tasks process asynchronously, so connections do not need to remain open. The API also provides automatic "reasoning summaries," making it easier to follow the model's logic. For customers with zero-data-retention agreements, OpenAI now allows encrypted reasoning data to be passed between API requests without storing it on their servers.
All these new features are available now through the Responses API. Pricing remains largely unchanged, using remote MCP servers does not incur extra fees and users are only billed for the output tokens generated.
