In this presentation, Udaiappa Ramachandran (Udai), CTO/CSO at Akumina and Microsoft MVP in Azure AI, introduces NLWeb, an open-source protocol and toolset from Microsoft that transforms any website into an AI-powered conversational experience. NLWeb leverages large language models and the Model Context Protocol (MCP) to handle natural-language queries, enabling dynamic, intelligent user interactions directly on the web. Udai walks through how NLWeb works, showcases real-world implementations from companies like Tripadvisor, Eventbrite, and O’Reilly, and shares how developers can get started using the GitHub repo and deploy to Azure. This talk, presented at Nashua CLOUD .NET and DevBoston Meetups, is a must-watch for developers, product teams, and AI enthusiasts exploring the future of human-centric web interfaces. Watch the full video here: https://youtu.be/x_4gbzUrW0s

1. Udaiappa Ramachandran (Udai)
CTO/CSO @ Akumina Inc., | Microsoft MVP
Web: https://udai.io
LinkedIn: https://linkedin.com/in/udair
Meetup: https://meetup.com/nashuaug
NLWeb

2. Agenda
 What is NLWeb?
 How It Works?
 Real-World Use Cases
 Model Context Protocol (MCP)
 Getting started with NLWeb

3. What is NLWeb?
 NLWeb = Natural Language Web, an open-source protocol + toolset by
Microsoft
 Turns any website into an AI-powered conversational app
 Implements a lightweight REST/MCP server that handles natural-
language queries and returns responses in Schema.org JSON
 Key Feature(s)
 The power of LLM on every website
 Searches over your item catalog
 Open standard – easily switch out the model, embedding model and vector
database
 Tools to import structured data
 MCP integration, so makes your website agent-compatible

4. Real-World Use Cases
 Tripadvisor: plan family trips conversationally.
https://techcommunity.microsoft.com/blog/azure-ai-services-blog/nlweb-p
ioneer-qa-tripadvisor/4415289
 Eventbrite: search events with internet-based queries.
https://techcommunity.microsoft.com/blog/azure-ai-services-blog/nlweb-p
ioneer-qa-eventbrite/4415314
 O’Reilly Media, Delish, Shopify, Qdrant, etc., are early adopters.
https://techcommunity.microsoft.com/blog/azure-ai-services-blog/nlweb-p
ioneers-success-stories--use-cases/4417070

5. Life of a Query

6. Model Context Protocol (MCP)
https://modelcontextprotocol.io/
introduction
1. Host with MCP Client (Claude, IDEs, Tools)
• This is where the question originates.
• A tool (e.g., CLI app, chatbot, or IDE plugin) with an
MCP Client sends the weather request.
2. MCP Protocol → MCP Server C
• The Host routes the request to MCP Server C using
the MCP Protocol.
• Why Server C? Because this agent is configured to
access Web APIs (external services) – perfect for
weather.
3. MCP Server C → Remote Service C (Internet)
• MCP Server C parses the request (“Boston” as the
location) and makes an external web API call to a
weather provider (e.g., OpenWeatherMap).
4. Remote Service C → MCP Server C
• The weather service returns data like:
"Boston, 54°F, Partly Cloudy"
5. MCP Server C → Host with MCP Client
• The result is sent back via the MCP Protocol.
6. Host Displays Result to User
• The original tool (IDE, CLI, chatbot) shows:
“The weather in Boston is 54°F and partly cloudy.”

7. Getting Started with NLWeb
 GitHub Repository: https://github.com/microsoft/NLWeb
 Get our sample running locally: docs/nlweb-hello-world.md
 Change your default model, vector DB, etc.: config directory
 Load data: docs/database-load-tool.md
 Understanding the “site” scope
 Query data in a sample frontend:
 http://localhost:8000/static/debug.html
 (and http://localhost:8000/static/str_chat.html)
 Deploy to Azure: docs/setup-azure.md
 YouTube: https://build.microsoft.com/en-US/sessions/BRK240
 https://news.microsoft.com/source/features/company-news/introducing-nlw
eb-bringing-conversational-interfaces-directly-to-the-web/

8. Thank you for your time and trust!
Nashua CLOUD .NET & Dev Boston