# NLWeb

> Turns a website into a conversational, queryable endpoint by combining Schema.org data, a vector store and an LLM — and every NLWeb endpoint is also an MCP server.

_The Agent Protocol Atlas · /protocols/nlweb · [JSON](/api/protocols/nlweb) · [all The Agent Protocol Atlas](/protocols)_

- **name:** NLWeb
- **full_name:** Natural Language Web
- **layer:** discovery
- **creator:** Microsoft (R.V. Guha)
- **status:** emerging
- **year:** 2025
- **one_liner:** Turns a website into a conversational, queryable endpoint by combining Schema.org data, a vector store and an LLM — and every NLWeb endpoint is also an MCP server.
- **spec_url:** https://github.com/nlweb-ai/NLWeb
- **snippet:**

```
POST /ask  { "query": "..." } → grounded natural-language answer + structured sources
```
- **abbreviation:** NLWeb
- **also_known_as:** Natural Language Web
- **canonical_spec_url:** https://github.com/nlweb-ai/NLWeb
- **entity_uri:** https://github.com/nlweb-ai/NLWeb
- **taxonomy_layer:** discovery
- **sub_layer:** conversational-endpoint
- **protocol_type:** declaration
- **central_problem:** Turns a website into a conversational, queryable endpoint so agents can ask natural-language questions and get grounded, source-backed answers.
- **maintainer:** Microsoft (project led by R.V. Guha) + community (nlweb-ai/NLWeb)
- **governance_body:** vendor (Microsoft) + open-source community
- **license:** MIT
- **maturity_tag:** emerging
- **current_spec_version:** — (verify-against-primary-at-build)
- **spec_date:** — (verify-against-primary-at-build)
- **launch_date:** 2025
- **last_verified:** 2026-06-15
- **transport:** HTTP (POST /ask); each NLWeb endpoint is also exposed as an MCP server
- **core_mechanism:** NLWeb combines a site's Schema.org structured data, a vector store, and an LLM behind a /ask endpoint; a query returns a grounded natural-language answer plus structured sources, and every NLWeb endpoint is simultaneously an MCP server.
- **discovery_endpoint:** POST /ask (and MCP server interface)
- **settlement_type:** —
- **adoption_metric:** — (verify-against-primary-at-build)
- **notable_adopters:** {"value":"Microsoft","source":"https://github.com/nlweb-ai/NLWeb"}
- **relationships:** {"predicate":"built_on","target":"schema-org","note":"NLWeb -built_on-> schema.org (research §2 seed triple). schema.org-for-agents is not a separate record in this 12-record pass; edge documented for the build-out."}, {"predicate":"complements","target":"mcp","note":"Every NLWeb endpoint is also an MCP server, so it pairs with the capability layer."}
- **ideal_use_case:** Giving a content-rich site a conversational, agent-queryable answer endpoint grounded in its own structured data.
- **when_to_use:** When you have Schema.org data and want agents to ask questions and get cited, grounded answers from your site.
- **when_not_to_use:** When a static curated link map (llms.txt) is enough, or when you need cross-agent coordination (A2A) rather than Q&A.
- **code_example:** POST /ask HTTP/1.1
Content-Type: application/json

{ "query": "What does Acme sell under $50?" }
// → { "answer": "...", "sources": [ { "url": "...", "schema": "Product" } ] }
- **source:** Project (Microsoft / R.V. Guha), built on schema.org, MCP-server duality: https://github.com/nlweb-ai/NLWeb ; seed triple research §2.
- **agent_readiness_link:** agent-readiness/discoverability
- **layer_legacy:** content