NLWeb

NLWeb is a protocol and reference implementation for adding natural-language interfaces to websites. It lets a site expose a structured endpoint that accepts natural-language questions and returns ranked results, summaries, or generated answers using the site's own data.

NLWeb is aimed at websites whose content can be represented as collections of structured items, such as products, articles, recipes, jobs, places, reviews, or events. It can use existing machine-readable data such as Schema.org, RSS, search indexes, and databases.

An NLWeb server commonly exposes:

• an /ask endpoint for natural-language queries;
• modes such as list, summarize, or generate;
• structured JSON responses with result items and metadata;
• streaming responses for interactive clients; and
• an MCP-compatible endpoint so agents can call the site as a tool.

This makes NLWeb adjacent to Retrieval-Augmented Generation (RAG) and real-time fact-checking, but its scope is narrower: it defines a standard way for a website to answer questions about its own public or permitted data.

NLWeb does not remove the need for source quality, authorization, or anti-abuse controls. A public NLWeb endpoint should define what data is exposed, how results are ranked, and how generated responses cite or preserve source provenance.

The NLWeb documentation describes NLWeb as a protocol for agents, applications, and humans to communicate with websites using natural language.