Search index

Where is it? Across 12,000 public sources.

A maintained index of the German public web, from the Bundestag down to individual municipalities: searchable by people in the browser and by agents over MCP.

In the making

The corpus

Around 12,000 official sources: parliaments, federal and state ministries, authorities, municipal councils, and the official pages of office holders. Not their private accounts and not general media; that line is drawn per host and nowhere else.

Two datasets

  • URL index: every page we fetch, with its host, when it was first seen and last fetched, its status and a hash of its content.
  • Content dataset: only the pages someone would actually read, such as Drucksachen, notices, council papers and speeches. An open-source page-type classifier tells them apart from listings, calendars, search result pages and login forms.

A page that changes is kept as a new version, so a document can be read as it stood on a given date.

One API over both datasets: full-text and hybrid search, facets, and documents as of a date. Every hit carries its provenance: the source, the URL, and when it was fetched. The same API serves two surfaces:

  • Web search, for people. In the making
  • A search tool over MCP, for agents. In the making

From a hit to live data

The index answers where is this? Where the source has an official API, the matching wrapper answers what does it say right now?, live and structured. And when a wrapper’s answer cites a document, the index retrieves it in full.

How it fits together: Two datasets and a facade →