NLWeb is a platform that enables natural language querying of structured data across the web. It allows users to ask plain-language questions and receive answers grounded in Schema.org structured data from websites, or have those questions translated into SQL queries against enterprise data sources like HubSpot or Jira.
The system can discover and route requests to other NLWeb agents published on the internet, and expose any NLWeb instance as an MCP server for direct querying by AI agents. It evaluates available language models on cost and quality to select the most suitable one for each request, and uses neural scoring to rank search results by relevance.