Release of NLWeb
Microsoft has potentially launched a technology poised to transform the entire internet with the release of NLWeb. Announced at Build 2025, this open-source project, seemingly simple, has the power to turn any website into an AI application in seconds.
What is NLWeb and Why It Matters
NLWeb is described as an open protocol and toolset designed to easily add a natural language conversation interface to websites. Its significance lies in its ability to allow not just human users, but also AI agents, to directly interact with websites. Microsoft CTO Kevin Scott reportedly referred to it as “HTML for the agent network”. This approach enables any website to join the burgeoning agent network without requiring massive technical overhauls.
Release of NLWeb Microsoft
How NLWeb Works?
The process behind NLWeb is quite innovative and leverages existing web infrastructure:
- Utilizing Existing Data Structures: NLWeb directly uses a website’s existing structured data, such as Schema.org markup and RSS feeds. This means publishers don’t need to completely rebuild their content infrastructure.
- Data Processing and Storage: The system includes tools to add structured data to vector databases, supporting semantic search and retrieval. It supports all major vector databases, offering developers flexibility.
- AI Enhancement Layer: Large language models enhance the stored data with external knowledge and context. For example, a user querying about a restaurant might receive information enriched with geographical data and reviews, providing a comprehensive, intelligent response.
The final output is a universal natural language interface serving both human visitors and AI agents, the latter accessing information programmatically through the MCP framework. NLWeb specifically focuses on enabling real-time interaction directly on the publisher’s website.
The main GitHub repository page for NLWeb (Natural Language Web), an open-source project by Microsoft aimed at enabling websites to become conversational AI platforms
Built on Standards: The Role of MCP
A key foundational element of NLWeb is Anthropic’s Model Control Protocol (MCP), which is increasingly seen as a de facto standard. MCP fundamentally connects AI systems to data sources. Microsoft views MCP as the transport layer, with NLWeb and MCP together providing the HTML and TCP/IP for the open agent network.
Forrester Senior Analyst Will McKeon-White highlights NLWeb’s distinct advantages, noting its ability to give AI systems better control over how they “see” website components. This allows for improved navigation and understanding of tools, potentially reducing errors in content interpretation and the need for interface redesigns.
Overview of the Model Context Protocol (MCP) architecture, illustrating how it enables Large Language Models (LLMs) and AI agents to securely connect with and utilize external tools, APIs, and data.
Navigating the Protocol Landscape: NLWeb vs. Others
The field of AI protocols is becoming crowded. NLWeb distinguishes itself from other initiatives:
- AIA (Agent-to-Agent): This protocol focuses on agent orchestration and communication, rather than AI-ifying existing websites. It uses defined patterns for structured task passing between agents, but its current implementation is tied to Google’s Gemini stack.
the Agent-to-Agent (A2A) protocol
- LLMs.txt: While superficially similar, LLMs.txt is fundamentally different. Its goal is to help large language models better access web content. It’s described less as an interactive tool and more like a web crawler attempting to infer intent or providing content in a structured format like markdown with training permissions. NLWeb, in contrast, focuses on real-time interaction on the publisher’s site.
Early Adopters See the Value of NLWeb
Microsoft isn’t just launching NLWeb; several organizations are already using it, including Chicago Public Media, Allrecipes, Eventbrite, Hearst (Delish), O’Reilly Media, Tripadvisor, and Shopify.
Andrew Odewahn, CTO of O’Reilly Media and an early adopter, sees the significant value. He states that:
“NLWeb leverages existing web best practices and standards developed over the past decade, making them available to large language models.”
Companies already optimizing metadata for SEO and marketing purposes can now use this rich data to make their internal AI smarter and more capable via NLWeb.
Odewahn emphasizes NLWeb’s value both for companies consuming public information (“What does this company do?”) and for those publishing private information. It offers an excellent way to expose internal information to internal LLMs without manual searching. He also notes that adopting NLWeb often isn’t a heavy lift, as many organizations already use the underlying standards. He suggests trying it now, as it’s open source, can run within existing infrastructure, and there’s “nothing to lose” but much to gain.
The Path to Adoption: Pilot or Wait?
Analysts offer varying perspectives on the immediate adoption timeline. Michael Ni of Constellation Research suggests:
“NLWeb is still in a very early stage, expecting substantial adoption only in 2-3 years.”
He sees it as a vision with potential but requiring ecosystem validation, implementation tools, and reference integrations before mainstream enterprise piloting. He recommends leading companies with specific needs consider piloting now.
Conversely, Maria Gorskikh of AIA advocates for an accelerated approach to avoid falling behind. She sees piloting NLWeb now as a smart and necessary step for companies with significant content or data. For these firms, it’s
“not a wait-and-see moment—more like early adoption of APIs or mobile apps”.
However, she advises caution for regulated industries like insurance, banking, and healthcare, suggesting they hold off on production use until neutral, decentralized verification and discovery systems (like the MIT NANDA project) are established.
The Dawn of the Agent Era
For enterprise AI leaders, NLWeb represents a watershed moment that should not be ignored. The core message is clear: AI will interact with your website, and you need to AI-ify it. NLWeb is presented as a particularly attractive way for publishers to do this. Just as RSS became a mandatory feature for websites in the early 2000s, users will soon expect this capability, anticipating the ability to search and find information, while AI agent systems will necessitate access to content. This is the promise of NLWeb.
Based on its GitHub repository, NLWeb is indeed a “deeply agnostic” platform — fully compatible across operating systems (Windows, macOS, Linux), vector databases (Qdrant, Snowflake, Milvus, Azure AI Search), and LLMs (OpenAI, Deepseek, Gemini, Anthropic, etc.).It’s both lightweight and scalable, capable of running on anything from cloud clusters to personal laptops — and soon, even mobile devices.
Community Reactions
Initial reactions shared online show both excitement and skepticism.
Hugo Alves (@ugo_alves) mentioned Sam Altman:
@sama Create an Ask based on this, allowing anyone to deploy ChatGPT as a web Q&A system.
Sam Woods (@sw00ds) sees it as game-changing:
If it’s as significant as it sounds, it could redefine web-based AI applications.
UdayKumar (@UdayKumar) sees transformative potential:
Yes. It has huge potential to turn websites into conversational agents. A reference implementation will accelerate adoption.
Of course, there are also skeptical voices.
Pavel Snajdr (@PavelSnajdr) expressed a different view:
I’d prefer an “AI as cheap as electricity” approach — then we wouldn’t need any MCP band-aids.
Conclusion:
The emergence of NLWeb signals that the web is transitioning towards the agent era. The question of how to interact with AI is no longer optional for every website; it will become a necessity.
Find more AI news here !
