Extending MCP Servers with Interactive User Interfaces

This article introduces the proposed MCP Apps Extension (SEP-1865), a new standard for supporting interactive user interfaces within the Model Context Protocol (MCP). This extension responds to a major community demand by building upon previous efforts like MCP-UI and OpenAI Apps SDK, enabling MCP servers to deliver rich, interactive UIs to hosts.

The MCP Apps Extension establishes a consistent method for declaring UI resources, associating them with tools, and facilitating two-way communication between embedded interfaces and the host application. Its development involved contributions from MCP Core Maintainers at OpenAI and Anthropic, along with creators and leaders from the MCP-UI community.

Current MCP capabilities mainly handle text and structured data exchanges. While sufficient for many tasks, this setup becomes limiting when visual information or complex user inputs are necessary. For example, a data visualization server might send JSON chart data, requiring the host to interpret and render it—a process that burdens client developers with custom rendering logic. As UI needs grow more complex, interactions like gathering multiple user settings become cumbersome, leading to awkward prompt-response exchanges.

To address these issues, the community has experimented with various ad-hoc solutions, resulting in inconsistent implementations that risk fragmenting the ecosystem. Standardization is crucial to promote interoperability and prevent ecosystem divergence.

The MCP-UI project, initiated by Ido Salomon and Liad Yosef, has pioneered delivering interactive interfaces as first-class MCP resources. Backed by a large community and supported by SDKs used by industry leaders like Postman, Shopify, Hugging Face, Goose, and ElevenLabs, it demonstrates the feasibility of agentic apps with rich UIs within MCP.

The OpenAI Apps SDK further proved the value of incorporating sophisticated UIs in conversational AI, allowing developers to create engaging, interactive experiences inside ChatGPT with MCP as its foundation. To ensure consistency, interoperability, and security, OpenAI, Anthropic, and MCP-UI collaborate on an official MCP extension for interactive interfaces.

The proposed MCP Apps Extension outlines a schema for UI resources, extending beyond simple specifications to serve as a foundation for a new agentic app runtime—enabling novel interactions between AI models, users, and applications. The initial design focuses on core patterns with promising plans for expansion.

A key feature is pre-declared UI resources, identified via the ui:// URI scheme, which are registered and referenced within tool metadata. This approach allows hosts to prefetch and review UI templates, enhancing performance and security, and separating static presentation from dynamic content. For example, a server registers a bar chart template with a unique URI, and tools reference it in their metadata, streamlining communication and rendering.

In conclusion, the MCP Apps Extension aims to make MCP a more versatile and standardized platform for rich, interactive AI applications. Its development signals a shift towards more engaging and seamless human-AI interactions, paving the way for innovative solutions across industries.

FAQs

Q: What is the MCP Apps Extension?
A: It is a proposed standard to add support for interactive user interfaces in the MCP, enabling MCP servers to deliver and manage rich UI resources.

Q: Why is this extension important?
A: It addresses current limitations in MCP’s data exchange, allowing visual, complex, and dynamic user interactions that improve usability and reduce developer burden.

Q: Who contributed to this proposal?
A: The extension was developed by MCP Core Maintainers at OpenAI and Anthropic, along with MCP-UI community creators and leaders.

Q: How will this impact developers?
A: Developers will be able to create more sophisticated, secure, and efficient AI applications with standard tools for deploying rich UI components within MCP.

Q: When will this extension be available?
A: The proposal is still in development; future implementation timelines depend on community feedback and organizational approval.