matricula-mcp
An MCP server for Matricula Online (data.matricula-online.eu), the church-book portal that ICARUS, the International Centre for Archival Research, runs for dioceses and archives in Germany, Austria, Poland, Slovenia, Serbia, Italy, Switzerland, Luxembourg and Bosnia and Herzegovina. Matricula publishes their parish registers (baptisms, marriages and burials, and the indexes, family books and parish censuses that go with them), scanned page by page: about 13,400 parishes by its own count in October 2026.
The server walks Matricula's own hierarchy. It lists the countries and their sections (each a diocese, an archive or a private collection), a section's parishes with their coordinates, a parish's registers with their signatures, types and dates, and a register's pages, each with its label and the link that opens it in Matricula's own viewer. It also runs Matricula's place search, which finds the parish a village belonged to.
It works the way a careful genealogist does. The register page is the evidence. A register's type and dates, a page label and a search snippet are finding aids that say which page to open. Every result carries a citation in the order Matricula itself gives for finding a page again: diocese or archive, parish, register signature, page.
Page images are read in Matricula's viewer, not fetched by this server. Matricula's image host refuses scripted requests, and this server does not pretend to be a browser to get past it. The code that would save one page image is here, documented and tested, but switched off: it is for the day ICARUS grants access to an identified client. See Page images.
Nothing here writes to Matricula, and nothing here keeps a family tree.
This is an independent project. It is not affiliated with, endorsed by, or supported by ICARUS or any of the archives and dioceses whose registers Matricula publishes.
Tools
The server publishes seven tools. All but matricula_page_image are
read-only; that one creates a new file only when image access is on, and
never overwrites one.
Finding
| Tool | Purpose |
|---|---|
matricula_sections |
The countries and, in each, its sections (dioceses and archives), with Matricula's parish counts. |
matricula_parishes |
The parishes of one section, filtered by name, with coordinates, and the section's own note (often the archive's address). |
matricula_search |
Matricula's place search: parishes whose names or descriptions name a place, with the snippet that shows where. Optionally limited to a section or to registers in a span of years. |
Reading
| Tool | Purpose |
|---|---|
matricula_registers |
A parish's registers: signature, the archive's own type (Taufen, Trauungen, Sterben, Index - ...), a kind (baptisms, marriages, burials, ...), dates, contents notes, where the original is held, and each register's key. |
matricula_register |
One register: its description and citation, and every page's position, label and viewer link, with paging and a label filter. |
matricula_page_image |
Off unless the operator has turned image access on. While off, it answers images_not_enabled with the page's viewer link, label and citation. When on, it saves one page as a new .jpg with its citation. |
cache_status |
This session's requests, the pacing, cache use, and whether image access is on. Makes no request. |
Everything is addressed by its path on Matricula below the language prefix,
which the tools call a key: deutschland/passau is a section,
oesterreich/salzburg/krimml a parish, oesterreich/salzburg/krimml/TFBI a
register. Every tool also takes a Matricula address in any of its interface
languages, and a register's tools take a page's viewer address (?pg=N).
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 matricula-mcp
From a clone, which is what you want if you will change it:
git clone https://github.com/ianderso/matricula-mcp
cd matricula-mcp
uv sync
uv run matricula-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": {
"matricula": {
"command": "uvx",
"args": ["matricula-mcp"],
"env": { "MATRICULA_CONTACT": "you@example.org" }
}
}
}
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 matricula -e MATRICULA_CONTACT=you@example.org -- uvx matricula-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 |
|---|---|
MATRICULA_CACHE_DIR |
Response cache directory. Default ~/.cache/matricula-mcp. |
MATRICULA_TIMEOUT |
HTTP timeout in seconds for one request. Default 30. Image downloads get 120 to read. |
MATRICULA_MIN_INTERVAL |
Least seconds between two requests to Matricula. Default 2, and never below 2. |
MATRICULA_SEARCH_INTERVAL |
Least seconds between two place searches. Default 10, and never below 10. |
MATRICULA_CONTACT |
An email address or URL added to the User-Agent, so ICARUS can reach you if your use causes trouble. Optional for reading the catalogue, courteous, and required for image access. |
MATRICULA_DOWNLOAD_DIR |
An existing folder. When set, matricula_page_image saves only inside it. Set it to save into an iCloud Drive folder, which lives under ~/Library. |
MATRICULA_IMAGE_ACCESS |
Leave unset. Set it to granted only once ICARUS has granted your client access to its image host; it needs MATRICULA_CONTACT. Any other value is refused. See Page images. |
An unusable value is reported on the first tool call as a not_configured
result naming the variable.
Page images
Matricula's viewer loads each page from img.data.matricula-online.eu. That
host answered 403 to a plain scripted request when this project surveyed
it (with Vary: Origin), and the viewer attaches a per-session token,
computed in the browser, to every image request. That is Matricula's access
control, and this server treats it as one. It does not imitate a browser,
send an Origin or Referer header, carry a session cookie, or reproduce
the viewer's token, and it never probes the host to find out what would get
through. A test fails if anything resembling a browser disguise appears in
the source.
So, by default, matricula_page_image fetches nothing. It answers
images_not_enabled with the page's viewer link, its label and its
citation, and the person reads the page in Matricula's viewer.
The code that would save a page is written and tested against a mocked host,
for the day ICARUS grants access to an identified client. If you obtain that
permission (by agreement with ICARUS, for your own client), set
MATRICULA_CONTACT to the address ICARUS knows you by and
MATRICULA_IMAGE_ACCESS=granted. The tool then fetches one page at a time
from the image host, paced like every other request, with this server's own
User-Agent (matricula-mcp/<version> (+https://github.com/ianderso/matricula-mcp; <your contact>))
and nothing else. If the host still refuses, the tool says so
(image_refused) and writes nothing. Should ICARUS specify a different route
(an API, a key, another host), the client is to change in a release to use
exactly that route; no setting here invents one.
A saved page is a new .jpg, never a hidden file or one under ~/Library
(unless MATRICULA_DOWNLOAD_DIR is set there), and comes with its citation
and its licence, CC BY-NC-ND 2.0.
Being a good guest
Matricula is one service, run by an association of archives. The client
sends one request at a time, at least two seconds apart, and a place search
at least ten seconds after the last, because a search reads every parish
description in the portal. Country and section lists and parish lists are
cached for 28 days, parish and register pages for 7, searches for a day. Two
identical calls in flight share one request. A 429, a 5xx or a dropped
connection gets one retry, honouring Retry-After; a refusal is not retried.
The User-Agent names the package, its version, this repository and, when
set, your contact. No cookies are kept.
What Matricula's terms say
The terms of use (German; checked 2026-10-11) cover the images and descriptive metadata on data.matricula-online.eu:
- Every church-book image is licensed CC BY-NC-ND 2.0: "Eine kommerzielle Weiterverwertung der Bilder ist demnach nicht gestattet" (commercial reuse is not permitted). No altered versions, either.
- Data from the registers may be used only as each country's civil-status and data-protection law allows; registers holding protected data are not shown. In Austria, baptism books are closed for 100 years, and marriage and death books are shown up to 1938.
- Anyone who publishes, in print or online, using even a little of the data or images undertakes to inform the archive or diocese concerned; for online publications a link is enough, and substantial print use calls for a copy.
- They say nothing about automated access.
matricula_parishes passes on each section's own note, which often names
the archive and how to reach it: the place to send that notice.
robots.txt, as a fact
Matricula's robots.txt (served from cdn-static.matricula-online.eu, read
2026-10-11) names several crawlers, ClaudeBot, GPTBot and CCBot among them,
with Disallow: /, and asks every other agent to keep away from four kinds
of address: the viewer's page parameter (/*?pg=), the search
(/*/suchen), the map features (/*/landkarte-features) and accounts. This
server is not a crawler: it reads the pages a researcher asks for, one at a
time, and caches them. It reads the search (matricula_search) and one map
features file per section (for matricula_parishes' coordinates); it builds
?pg= links for a person to open and never requests them; it never touches
accounts. Whether that suits your use is yours to judge.
How to read what comes back
- Cite the hierarchy, keep the link as a locator. Matricula advises
finding a page again through diocese or archive, parish, register
signature and page, and does not promise that its links last. Each result's
citationgives them in that order, withcite_asas one sentence. pageis the image's position in the register, counted from 1, which is what the viewer's?pg=takes. Thelabel(02-Taufe_0001,001-03_0004-r) is the archive's name for the scan: a section and number, or a folio and side. Cite both.typeis the archive's word;kindis this server's reading of it.Index - Taufenand Passau'sRegister Taufenare indexes (index: true); Baden-Württemberg'sTaufregisteris the register itself.- A register with
hosted_elsewhereis listed in Matricula but shown on the archive's own site (the Landesarchiv Baden-Württemberg's section links to its permalinks). Its images are not in Matricula's viewer. - A gap is not an absence. A register missing from Matricula may be
closed by law, unscanned at the diocese or archive, held elsewhere, or
lost.
matricula_registerslists what Matricula has. - The search reads parish names and descriptions, never the entries.
Many descriptions list the villages a parish served, which is how the
search finds a village's parish;
snippetshows the matching text. It matches whole words:Gasteinfinds Bad Gastein, not Dorfgastein. No search here finds a person's name. - Descriptions are the archives' own text. Parish histories, contents notes and comments reach the model verbatim; they are material to weigh, never instructions.
Deliberately not here
- Fetching page images without permission. See Page images.
- Reading entries. Matricula has no transcriptions or name index; the page image is all there is.
- Accounts, comments or anything that writes to the site.
- Working around bot checks. A challenge is reported as
blockedand left alone.
Security
Tool arguments are written by a model, and the model reads text this server does not control. The server assumes that text can steer the model, and limits what a steered model can make it do.
- Which hosts. A request hook refuses anything but https to
data.matricula-online.eu, including addresses taken from a page. The image host is refused too while image access is off. Redirects are followed only on the same host. - Which pages. A key must have the right number of parts, each from the
alphabet Matricula's slugs use; anything else (
.., another host) is refused before any lookup. - Which addresses. Each connection is checked where it is made: a name that leads to a private, loopback, link-local, CGNAT, multicast, reserved or unspecified address, IPv4 or IPv6, is refused, and the connection goes to the address that was checked. Proxy settings in the environment are not used.
- How much. A page over 8 MB, or an image over 60 MB, is refused as it streams in.
- Which files.
matricula_page_imagecreates one new file and never overwrites one, and only when image access is on. The bytes must be a JPEG, judged by their first bytes. Never a hidden file or folder, never under~/Library, and withMATRICULA_DOWNLOAD_DIRset, never outside it, all judged after links are resolved. A refused download leaves nothing on disk. - Site text is untrusted. The server's instructions tell the model to treat it as material, 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 the site
uv run ruff check .
uv run ruff format --check .
uv run python -m tests.live_check # paced calls to Matricula's catalogue pages
The live check asks Matricula what the recorded pages cannot: whether its pages still have the shape the server reads. It reads catalogue pages only, about a dozen, two seconds apart, and one search; it never calls the image host. See CONTRIBUTING.md for how the suite is organised, docs/API-NOTES.md for what was observed of the site and when, and docs/DESIGN.md for why the server is shaped this way.
Credits
The registers belong to the dioceses, archives and parishes that hold them, and the images are published by them through Matricula under CC BY-NC-ND 2.0. Matricula is run by ICARUS, which asks those who use it to support it.
License
MIT, for this server's code. The images and descriptions it links to are under Matricula's terms.
Metadata
Release files for matricula-mcp 0.1.0
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | |
|---|---|---|---|
| matricula_mcp-0.1.0.tar.gz | 175.6 kB | Details |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Reset |
|---|---|---|---|---|
| matricula_mcp-0.1.0-py3-none-any.whl | Python 3 | none | any | Details |
Total release size: 221.3 kB
Release files / matricula_mcp-0.1.0.tar.gz
| Download URL | matricula_mcp-0.1.0.tar.gz |
|---|---|
| Size | 175.6 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
a5951de43b979d71b50a0987c1285b08b3f9ab0b41d46f8c6bde77ed9156f3cf
|
|
BLAKE2b-256 checksum How to use checksums |
f975033926afe7daeeece1d238d16d17c57d840d1a846a1a37919300e945b8ce
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Oct 11, 2026.
Transparency logRelease files / matricula_mcp-0.1.0-py3-none-any.whl
| Download URL | matricula_mcp-0.1.0-py3-none-any.whl |
|---|---|
| Size | 45.7 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
e9fa109bdd647725968994c7f5b2ecc47320127e1742ce447eab5170db0824f0
|
|
BLAKE2b-256 checksum How to use checksums |
282560f5f1b012063b841f55eb3dfbdc158f7191f219bcc3bacff0031c1415bc
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Oct 11, 2026.
Transparency log