Gov Gazetteer

CI PyPI

An MCP server for GOV, the Geschichtliches Orts-Verzeichnis: the historical gazetteer of the Verein für Computergenealogie (CompGen), with about 1.2 million places, mostly in Germany and the lands that were once German or Prussian. GOV dates everything it records about a place: its names, what kind of place it was, and every place that enclosed it, civil and ecclesiastical.

That makes it the tool for the question a genealogist asks of every German village: in the year of the event, which Kreis, province and state held it, and which parish and Standesamt kept its registers? A village in Posen was Prussian until 1919 and Polish after 1920; a Pomeranian one was German until 1945. A Kreis was renamed, an Amt dissolved, a parish merged. The server reads GOV's public web service and answers for one year at a time, and it gives the dated enclosures a genealogy program needs to build a place tree that is right for every date.

It works the way a careful genealogist does. A gazetteer entry is a finding aid, not evidence. GOV is edited by volunteers: a parish or Standesamt link says which register to look in, and the register's own heading confirms it. Every result credits GOV and CompGen and carries the address of each object's page on GOV, because that page is what gets cited.

Nothing here writes anywhere, and nothing here keeps a family tree. It sits well beside familysearch-mcp, whose place tools resolve a place by date but know no parishes or Standesämter, and us-places-mcp, its counterpart for American counties.

This is an independent project. It is not affiliated with, endorsed by, or supported by the Verein für Computergenealogie.

Tools

The server publishes six tools. All are read-only.

Tool	Purpose
gov_search	Find places by name: GOV id, names, types, coordinates and current parents. within limits the search to what lies inside a Kreis, province or state at any date; place_type keeps villages, parishes, Standesämter and so on.
gov_object	One object in full: every name with its language and dates, every type with dates, coordinates, external ids, and each part-of, located-in and represents relation with its dates and GOV's source ids. children lists what lies inside it, such as a parish's villages. Takes a GOV id, a GOV page address, or an external id such as geonames:2851465.
gov_place_at_date	A place in one year: its name then and the civil chain above it to the state, each level with its type and the link's dates. Links GOV leaves undated, or attests only in another year, are marked.
gov_registers	The parish (by confession, where GOV records it) and the Standesamt that covered a place in one year, the bodies above each parish whose archive may hold the books, and the local court. Says plainly when GOV records no parish or no Standesamt.
gov_enclosures	Every dated enclosure of a place and of each civil place above it, shaped for a dated place tree: one entry per parent with its periods, current first, each with a gramps_date such as from 1871-01-01 to 1937-03-31, and a suggested Gramps place type.
cache_status	This session's requests to GOV, by operation, and cache use. Makes no request.

Setup

You need Python 3.11 or later and uv. There is no key to request.

Without cloning. uvx fetches it from PyPI and runs it in one step:

uvx gov-gazetteer-mcp

From a clone, which is what you want if you will change it:

git clone https://github.com/ianderso/gov-gazetteer-mcp
cd gov-gazetteer-mcp
uv sync
uv run gov-gazetteer-mcp   # stdio server, usually launched by the client

Either way the server speaks MCP over stdio, so you will normally let an MCP client start it rather than run it by hand.

Claude Desktop

{
  "mcpServers": {
    "gov": {
      "command": "uvx",
      "args": ["gov-gazetteer-mcp"]
    }
  }
}

A desktop app does not always inherit your shell's PATH. If the server fails to start because uvx cannot be found, give the full path that which uvx prints as the command.

Claude Code

claude mcp add gov -- uvx gov-gazetteer-mcp

Configuration

Nothing is required. A .env file in the directory the server starts in supplies anything the environment does not; only that directory is read.

Variable	Meaning
GOV_GAZETTEER_CACHE_DIR	Response cache directory. Default ~/.cache/gov-gazetteer-mcp.
GOV_GAZETTEER_TIMEOUT	HTTP timeout in seconds for one request. Default 30.
GOV_GAZETTEER_MIN_INTERVAL	Least seconds between two requests to GOV. Default 1, and never below 1.
GOV_GAZETTEER_CONTACT	An email address or URL added to the User-Agent, so CompGen can reach you if your use causes trouble. Optional, and courteous.

An unusable value is reported on the first tool call as a not_configured result naming the variable. The server writes no files but its cache.

Being a good guest

GOV is run by volunteers on CompGen's own server. The client sends one request at a time, at least a second apart, and two identical calls in flight share one request. Objects are cached for 30 days and searches for 7, and every object an answer carries (a search returns whole objects, not ids) is kept for the session, so a place already seen is not asked for again. A question about one year reads each enclosing object once; the same place in another year then costs nothing. A walk reads at most 40 objects per call and says so if it stops. A 429, a 5xx or a dropped connection gets one retry, honouring Retry-After; a SOAP fault is GOV's answer and is not retried. The User-Agent names the package, its version and this repository.

Terms of use

GOV's site has no terms-of-use page; its footer links CompGen's Impressum and privacy statement. The home page says the project was started to make "qualitativ hochwertige Daten für jedermann bereitzustellen" (to provide high-quality data for everyone, gov.genealogy.net), and it publishes its SOAP services without a key at gov.genealogy.net/services. CompGen describes GOV's data as freely available Linked Open Data; GOV's documentation on GenWiki sits behind a bot check and was not read for this project. Every result credits "GOV — Geschichtliches Orts-Verzeichnis, Verein für Computergenealogie (CompGen)" and links each object's page.

As a fact: GOV's robots.txt disallows a few website paths (edit and history pages, KML and GeoJSON exports, the distance search) and names several crawlers it shuts out entirely; it does not mention /services/, the web service this server uses.

How to read what comes back

• Ask with the event's year. gov_place_at_date and gov_registers answer for one year. A place that changed Kreis or state within that year shows both links with their dates; choose by the event's exact date.
• A parish or Standesamt link is a lead. GOV is edited by volunteers. Confirm the parish or office in the register itself, whose heading or title page names it and the places it served, and cite the register.
• Certainty is marked. A link is dated (its dates bracket the year), undated (GOV gives none; it is assumed to hold), or attested in another year only (GOV records it from one directory of another year). Prefer the first.
• Where GOV links no parish to a place, gov_registers gives the parishes whose church stands in the place, and says so: usually, not always, the place's own parish. Where GOV links no Standesamt, it gives a Standesamt of the same name in the same district, as a candidate, and says GOV does not link them. In well-modelled regions (Pomerania, for one) GOV links both directly.
• Name search is literal. It matches whole words and word beginnings, ignoring case and accents (Lubeck finds Lübeck), and never spelling variants: German, Polish and older spellings (Cöslin, Köslin, Koszalin) each need their own call. A zero covers only the spellings and area searched. GOV answers at most 500 objects per search.
• Names follow the year. A Polish name GOV dates from 1945 is not given for 1900. Among names that hold, language (deu by default, pol for a place under Polish rule) picks the one shown; names_at_date lists them all.
• Before 1874 there was no Standesamt in Prussia (1876 in the rest of the Empire; earlier on the left bank of the Rhine). gov_registers says so for those years.
• Confederations are left out. GOV records memberships of the German Confederation, the Rheinbund, the League of Nations, the EU and the UN as enclosures; they are not levels of government, so a chain stops below them and a note names them.
• Cite the object's page. Use the citation each result carries: the source credit, the entry's title, its GOV id and address, the day GOV last changed it and the day it was read.

Deliberately not here

• Editing GOV. GOV's ChangeService needs an account; this server builds no envelope for it.
• The GenWiki documentation. It sits behind a bot check (Anubis), which this project does not work around. Everything here was built from the WSDLs and GOV's live answers; see docs/API-NOTES.md.
• Bulk export. This server reads places one at a time for research, not whole regions.

Security

Tool arguments are written by a model, and the model reads text this server does not control, including GOV's own names and notes. The server assumes that text can steer the model, and limits what a steered model can make it do.

• One address. Every request is a POST over https to gov.genealogy.net/services/ComplexService; a request hook refuses anything else, and an argument chooses only what is asked, never where. Redirects are not followed.
• Read operations only. The client builds envelopes for seven read operations and nothing else.
• Public addresses only. The connection goes to an address checked to be public, so a DNS answer pointing at a private network is refused. Proxy settings in the environment are not used.
• Bounded, plain answers. An answer over 10 MB is refused as it streams in, and one carrying a document type declaration is refused before it is parsed.
• Arguments are validated (GOV ids, external ids, years, type names) before they reach a request.
• GOV's text is untrusted. Names and notes reach the model verbatim. The server's instructions tell the model to treat that text as material to weigh, never as instructions; the model still decides, so review what it proposes to do.

To report a vulnerability, see SECURITY.md.

Development

uv sync --extra dev
uv run pytest                      # mocked with respx; never touches GOV
uv run ruff check .
uv run ruff format --check .
uv run python -m tests.live_check  # paced calls to GOV itself

The live check asks GOV what the recorded fixtures cannot: whether its answers still have the shape the server reads. See CONTRIBUTING.md for how the suite is organised, docs/API-NOTES.md for what was observed of the service and when, and docs/DESIGN.md for why the server is shaped this way.

Credits

The data is GOV's — Geschichtliches Orts-Verzeichnis, a project of the Verein für Computergenealogie e.V. (CompGen), built by its volunteers. GOV's software is by Jesper Zedlitz.

License

MIT.