Inblog
Start your company blog
GEO

NLWeb

NLWeb (Natural Language Web) is an open protocol and open-source project Microsoft announced at Build in May 2025. It gives any website a natural language interface, and every NLWeb instance doubles as a Model Context Protocol (MCP) server. The project is led at Microsoft by R.V. Guha — creator of RSS, RDF, and Schema.org — and is often summed up as "NLWeb is to MCP what HTML is to HTTP."

NLWeb (Natural Language Web) is an open protocol and open-source project Microsoft announced at Build in May 2025. It gives any website a natural language interface, and every NLWeb instance doubles as a Model Context Protocol (MCP) server. The project is led at Microsoft by R.V. Guha — creator of RSS, RDF, and Schema.org — and is often summed up as "NLWeb is to MCP what HTML is to HTTP."

Why It Matters

Until now, a website's main path into AI systems was passive: crawlers collected pages and carried them elsewhere. NLWeb flips that, turning the site itself into an active endpoint that receives questions and answers them. Visitors search the site's content in natural language, while AI agents call the same interface over MCP to query site data directly. Humans and agents sharing one natural language API is the core idea — and it is how individual sites stay first-class participants in the agentic web.

How It Works

NLWeb feeds on the semi-structured data sites already publish.

  • Data layer: Existing Schema.org markup, RSS feeds, and similar structured data are collected and indexed into a vector store.
  • The ask method: The core API takes a natural language question, processes it with an LLM, and returns the answer as Schema.org-formatted JSON.
  • MCP server: Because every NLWeb instance is also an MCP server, MCP-compatible AI clients and agents can query the site without any custom integration.

Adoption

Launch partners included Tripadvisor, Shopify, Snowflake, O'Reilly Media, and Eventbrite. In August 2025, Cloudflare combined NLWeb with its AutoRAG infrastructure: pick a domain, and crawling, indexing, and deployment happen automatically. That managed path means a site can become conversational in a few clicks, without running its own server. Build 2026 was the first stage where the protocol was judged on documented deployments rather than potential.

GEO Implications

NLWeb multiplies the return on structured data. Schema.org markup stops being extra information for search engines and becomes the raw material for a conversational interface and an agent API. And where llms.txt is a brochure telling AI what content exists, NLWeb is the execution layer that answers questions directly. Beyond being cited by external AI search, a site gains its own channel for answering from its own content.

Sources:

How inblog Helps

Protocols like NLWeb ultimately depend on the quality of the structured data a site publishes. inblog blogs ship with clean markup and metadata structure by default, so a conversational layer like NLWeb can index the content as-is. Keep headings clear and the body well structured, and turning your site into a queryable node of the agentic web requires minimal extra work.