What is NLWeb?

https://media2.dev.to/dynamic/image/width=1080,height=1080,fit=cover,gravity=auto,format=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2Fdmn6p54zkwh5afse2kag.jpg

Imagine visiting a website and, instead of scrolling, clicking, or digging through menus, you simply ask it a question and get exactly what you need. That’s the promise of NLWeb, an open project from Microsoft designed to make natural language interfaces a standard part of the web experience.

In simpler terms, NLWeb makes it easy to turn any site into an AI-powered app.

So, what exactly does NLWeb do?

NLWeb allows websites to integrate conversational AI features directly into their pages. This means that instead of manually searching or navigating, users can query a site’s content in plain language, much like chatting with an AI assistant or a co-pilot.

For example:

• On a news site, you could ask, “Show me all articles about renewable energy published in the last 6 months.”
• On an e-commerce site, you might say, “Find me red sneakers under $80 with free shipping.”
• On a university page, you could ask, “Which Master’s programs have a December application deadline?”

No more endless clicking, just ask, and the AI responds.

The Key Components of NLWeb

1. Natural Language Interfaces for Any Site

NLWeb gives site owners a straightforward way to integrate AI models of their choice (OpenAI, Azure AI, local models, etc.) into their sites. They can also hook it up to their data, so the AI’s responses are relevant, accurate, and up-to-date.

2. MCP (Model Context Protocol) Integration

Every NLWeb instance is also an MCP server, meaning it can expose a site’s content in a structured way to other AI agents or participants on the web.

Think of MCP like a universal translator for AI agents: it lets one AI-powered site talk to another, share relevant data, and even collaborate.

3. A Step Towards the Agentic Web

Microsoft sees NLWeb as playing a role similar to HTML, but in the emerging agentic web. Just as HTML made websites readable by browsers, NLWeb aims to make web content readable and actionable by AI agents.

How It Works

NLWeb uses semi-structured formats like:

• Schema.org offers standardized metadata for web content.
• RSS Feeds provide structured updates for blogs, news, and podcasts.
• Other published data — product listings, event calendars, knowledge graphs, etc.

These are combined with LLM-powered tools to understand and respond to user queries in natural language.

Flow of NLWeb + MCP

Here’s a simplified visual of how NLWeb works in combination with MCP:

1. User asks a question.
2. NLWeb passes it to the AI model.
3. The model fetches answers from the site’s data.
4. MCP can share this content with other AI agents if enabled.

Example: Setting Up NLWeb + MCP

Here’s a minimal JavaScript example to illustrate how NLWeb might be set up:

import { NLWebServer } from "@microsoft/nlweb";

// Create a new NLWeb server with MCP support
const server = new NLWebServer({
  model: "gpt-4", // or your preferred LLM
  dataSources: [
    { type: "rss", url: "https://example.com/feed" },
    { type: "schema", url: "https://example.com/schema.json" }
  ],
  mcp: {
    enabled: true, 
    endpoint: "https://example.com/mcp"
  }
});

// Start the server
server.listen(3000, () => {
  console.log("NLWeb server running at http://localhost:3000");
});

// Users (or AI agents) can now ask:
// "Show me the latest blog posts about AI from this site"

Why NLWeb Matters

• Better User Experience → No more hunting for the right menu or search keyword.
• Customizable AI → Site owners choose the model and training data.
• Interconnected Web → MCP means AI agents can discover and access information across multiple sites.
• Future-Proof → As AI agents become more common, NLWeb could be the HTML of the AI era.

The Bigger Picture

We’re moving towards a web where AI is not just a feature, it’s the interface. Instead of websites being passive collections of pages, they become active participants that can:

• Answer questions
• Offer recommendations
• Perform tasks
• Share information with other AI agents

In this vision, NLWeb is a building block of the “Agentic Web,” a future where websites, AIs, and users interact fluidly through conversation.

Getting Started…

Microsoft has made NLWeb open-source and available on GitHub. Developers can start experimenting, connect their own AI models, and even enable MCP to let their site’s data become discoverable to other AI agents.

If this vision takes off, we might look back at NLWeb the way we look at HTML today as the foundation of a new era of the web. The next time you visit a site, imagine just asking it what you need and it answering instantly. That’s the future NLWeb is aiming to make real.

References:

https://news.microsoft.com/source/features/company-news/introducing-nlweb-bringing-conversational-interfaces-directly-to-the-web

https://github.com/microsoft/NLWeb