Microsoft, along with its partners, is embarking on a groundbreaking journey to revolutionize web search through the integration of natural language interfaces. At the core of this innovation lies the Model Context Protocol (MCP), a powerful tool designed to cater to both human users and artificial intelligence agents alike.
Imagine a world where interacting with websites and web applications becomes as intuitive as having a conversation. This vision is becoming increasingly tangible thanks to technologies like MCP that are reshaping the landscape of AI-powered applications. By establishing a standardized approach to describing and accessing information within applications, MCP simplifies the process of incorporating AI capabilities into existing software and services.
As Microsoft CTO Kevin Scott aptly puts it, we are witnessing the dawn of “the agentic web,
” with MCP poised to become the foundational framework for interconnected AI applications – akin to what HTTP represents for the internet infrastructure.
The advent of NLWeb signifies a pivotal moment in this technological evolution. Unveiled at Build 2025, NLWeb is an ambitious open project aimed at standardizing the development of natural language interfaces for websites and web applications. Empowered by cutting-edge generative AI models, NLWeb enables voice and text inputs that transcend traditional click-based navigation, opening up new horizons for accessibility and seamless machine-to-machine interactions.
One key aspect setting NLWeb apart is its platform-agnostic design. Compatible across various operating systems and seamlessly integrating with common models and data tools, NLWeb leverages the semi-structured semantic foundation of the web ecosystem. From basic HTML document structures to advanced technologies like Schema.org and micro-formats, NLWeb harnesses these building blocks to create a contextual semantic search layer that transcends conventional boundaries.
At its core, NLWeb operates as an MCP server within each implementation, enabling seamless integration into larger-scale agent orchestrations as part of dynamically composed applications. The collaborative efforts between Microsoft and its partners have culminated in an open development initiative on GitHub – offering not just code snippets but also a comprehensive protocol utilizing JSON over REST for natural language query processing.
In practical terms, NLWeb excels when deployed on websites structured around “
lists of items” such as recipes, events, or places that leverage Schema.org microdata embedded in HTML content. Leveraging sophisticated AI algorithms trained on web content allows NLWeb to navigate these structured datasets efficiently without necessitating bespoke algorithms for individual sites – making deployment cost-effective and user-friendly.
One standout feature of NLWeb is its extensible tool support system – enabling dynamic loading of processing features tailored to specific user queries based on predefined schemas. These tools span from default search functionalities to specialized item details retrieval tools or comparison utilities – empowering users with diverse interactive capabilities while ensuring flexibility in addressing unique use cases.
