TLDR: MCP-UI is a new project that aims to replace traditional websites with AI Agent User Interfaces, suggesting a shift towards more interactive and intelligent web experiences powered by AI agents. This initiative, driven by the Model Context Protocol (MCP), envisions a future where ‘Jarvis clients’ and MCP servers supersede conventional web browsers and websites, enabling rich, dynamic user interfaces within AI agent processes.
The digital landscape is on the cusp of a significant transformation as AI agents and the Model Context Protocol (MCP) are poised to redefine user interaction, potentially rendering traditional websites as ‘old world UIs.’ This week, at Microsoft’s MCP Dev Days, Kent C. Dodd articulated a bold vision for the future, stating, ‘we’re entering a new phase where, instead of web browsers and websites, we’re going to have Jarvis clients [a reference to AI agents] and MCP servers.’
While this shift doesn’t imply the obsolescence of web technology itself, it suggests a fundamental change in how users will engage with online content. The prevailing idea is that AI agents will continue to leverage the web as an interface layer for end-users. However, instead of navigating conventional websites, users will interact with new chat-based AI browsers, currently under development by entities like OpenAI and Perplexity, or through other programming methods that inject user interfaces directly into AI agent processes. This is precisely where the MCP-UI project comes into play.
MCP-UI empowers developers to create interactive UI components specifically for MCP. The project’s core objective is to ‘standardize how models and tools can request the display of rich HTML interfaces within a client application.’ This functionality is akin to how smartphone platforms such as iOS and Android enable developers to embed WebViews within their applications. More precisely, MCP-UI facilitates ‘rich, dynamic user interfaces for your MCP applications,’ currently offering SDKs for TypeScript and Ruby.
Still in its nascent and experimental stages, MCP-UI’s philosophy aligns seamlessly with the burgeoning popularity of MCP, which has become the de facto method for AI models to gather external information. The MCP-UI project asserts that ‘Allowing MCP servers to respond with UI snippets is a powerful way to create interactive experiences in hosts.’
During his presentation at Microsoft’s MCP event, Kent Dodd showcased MCP-UI in action, demonstrating an e-commerce scenario where a user interface was displayed within VS Code. Dodd enthusiastically declared, ‘This is the future of user interaction, right here. You’re no longer going to have a website that users go directly to, or anything like that. Now we are going to be interacting with a UI, still, but it’s going to be in the context of our AI agent that’s going to solve these problems for us. And I just love that.’
MCP-UI is an open-source initiative, licensed under Apache-2.0, and was developed by Israeli developers Liad Yosef and Ido Salomon. Their recent recruitment by enterprise IT company Monday.com for ‘several million dollars each’ underscores the intense demand for top-tier AI talent in the current market. While the project itself does not appear to be part of the Monday.com acquisition, its future trajectory under the creators’ new employment will be closely watched.
The origins of MCP-UI can be traced back to the MCP Community Working Group, led by Ola Hungerford and Tadas Antanavicius, both maintainers of the official MCP project, which was originally created by Anthropic.
Liad Yosef, one of MCP-UI’s creators, previously commented on the ‘old world UIs,’ stating, ‘In the near term we will see more and more tools like NLWeb, browser rendering and browser agents that will try to automate using the “old world” UIs, which are originally meant for humans. This makes sense as it immediately bridges the gap between AI capabilities and existing UI. The smart bet for the future will be on reversing this flow – turning these services to be MCP-first, LLM-consumable, and having a very lean human UI built *on top* of that. Services that will quickly expose themselves in this “reverse” way will benefit from an immediate and great agent compatibility, while still catering for the (gradually decreasing) traffic of human visitors.’
NLWeb, a Microsoft project unveiled in May, aims to transform websites into AI applications, enabling users to query site content using natural language, similar to an AI assistant or Copilot. While acknowledging concerns about AI’s impact on the human-centered web, the development of MCP-UI and similar projects highlights a crucial need for AI agents to present user interfaces more effectively to human users. Google Chrome senior engineer Paul Kinlan has echoed this sentiment, suggesting that dynamically generated UIs from LLM prompts will be a near-future reality, with web technologies like HTML, CSS, and JavaScript being ideally positioned to power these new interfaces.
Also Read:
- Oracle Integrates Model Context Protocol for Enhanced AI Agent Database Access
- AWS Enhances GitHub Workflows with Generative AI Integration via Amazon Bedrock and MCP
It is already common for AI agents to browse the web for information, utilizing headless browsers like Browserbase. OpenAI’s o3 model, for instance, visibly ‘thinks’ as it browses websites in the background. The emergence of custom web UIs generated as part of an AI agent’s output represents a novel and exciting frontier in this evolving landscape, with MCP-UI positioned as a key project to monitor.
