bundesregierung-mcp
# bundesregierung-mcp π©πͺ
mcp-name: io.github.mohammad-abbasi-sh/bundesregierung-mcp
Unofficial **read-only** MCP server for **[bundesregierung.de](https://www.bundesregierung.de)** β the German federal government's official site β via its listing pages (press releases, articles / Meldungen, speeches & press conferences) and document pages, whose full text is read from the server-rendered reading area. Latest press releases (612 at last check), articles (732) and speeches (567), newest first, keyword search across the whole site, and full document text β some press releases also carry the official English translation inline. No key, no login.
> Built by Mohammad Abbasi as part of the "Germany MCP factory" (Sepehr-style pipeline). Not affiliated with the German federal government or the Presse- und Informationsamt der Bundesregierung.
## For AI agents
Add to your MCP client config (stdio):
```json
{
"mcpServers": {
"bundesregierung": {
"command": "uvx",
"args": ["bundesregierung-mcp==0.1.0"]
}
}
}
```
Or run locally: `python -m bundesregierung_mcp.server` (needs `mcp<2`, `httpx`).
## Tools
| Tool | Description |
|------|-------------|
| `breg_press_releases` | Latest press releases (Pressemitteilungen) of the German federal government, newest first β `limit` 1..50, `page` 1-based. Items carry title, topline, link, type, publish time (press-release teasers carry no description) |
| `breg_articles` | Latest articles / Meldungen, newest first β `limit` 1..50, `page` 1-based. Items carry title, topline, link, publish time and a short description |
| `breg_speeches` | Latest speeches, press statements and press conferences (Pressekonferenzen und Reden), newest first β mostly full transcripts ("Im Wortlaut") β `limit` 1..50, `page` 1-based |
| `breg_search` | Search bundesregierung.de by keyword across the whole site (press releases, articles, speeches) β `query` (e.g. "Klima"), `limit` 1..50, `page` 1-based. Items link via a /breg-de/suche/ alias URL |
| `breg_get_document` | Full text of one document β press release, article or speech/transcript β `url`: a www.bundesregierung.de document URL (/breg-de/aktuelles/... or a /breg-de/suche/ alias). Returns title, publish time, kind, the German body + word count, and `body_en` with the official English translation when the site carries one inline |
| `breg_health` | Reachability probe (ok / blocked / error) β one polite request |
Example output (`breg_press_releases`, `limit` 1):
```json
{"section": "pressemitteilungen", "total_results": 612, "page_count": 62, "page": 1, "count": 1,
"items": [
{"title": "ErklΓ€rung von Bundeskanzler Friedrich Merz zum MehrjΓ€hrigen Finanzrahmen 2028-2034",
"topline": null,
"link": "https://www.bundesregierung.de/breg-de/aktuelles/pressemitteilungen/erklaerung-von-bundeskanzler-friedrich-merz-zum-mehrjaehrigen-finanzrahmen-2028-2034-2462246",
"type": "Pressemitteilung",
"published": "2026-11-21T12:25:00Z",
"description": null}]}
```
Notes:
- The server reads the site's **listing pages** (`/breg-de/aktuelles/{pressemitteilungen,artikel,pressekonferenzen-und-reden}`, verified live 2026-11-22). The results are embedded server-side as `BPA.initialSearchResultsJson` in the HTML β there is **no RSS feed**: `/breg-de/service/rss-feeds` 404s and the pages carry no `<link rel="alternate">` feed.
- **Pagination quirk, verified live:** the site's `?page=` is 0-based β `?page=N` returns page N+1. Page 1 carries 11 items, later pages 10. Listing tools fetch successive pages until the requested `limit` is met.
- **Document text, honestly:** the full text is read from the server-rendered `#rs_reading_area_content` paragraphs (statement: 7 German + 8 English paragraphs incl. heading, 249 / 306 words; speech transcript: 20 paragraphs, 1,113 words; article: 11 paragraphs, 577 words β all sampled live). Article pages mix in an image caption + "Foto:" credit line and trailing related-link lines ("Pressemitteilung des BMDS: ..."); all three are filtered out of the body.
- **English translations:** some press releases carry the official English translation inline (`lang="en-GB"` spans wrapping whole paragraphs) β it is returned separately as `body_en` (`has_english_translation: true`). Inline foreign words (e.g. "EUDI-Wallet") are also marked `lang="en-GB"` but are single words, so a segment counts as English only when the en-GB spans cover β₯ 50% of its text.
- **Speech dates, honestly:** speech/transcript pages ("Mitschrift Pressekonferenz") carry NO `<time datetime>` β their date is read from the header's German date line ("Freitag, 20. November 2026" β `2026-11-20`). The featured (image) teaser on the article listing likewise carries no `<time>` β its date falls back to the item's `sortDate` field (site-local time, no offset).
- Data: bundesregierung.de public listing and document pages β no key, no login, read-only.
## Development
```bash
python3 -m venv .venv && .venv/bin/pip install "mcp<2" httpx pytest
.venv/bin/pip install -e .
.venv/bin/python -m pytest tests/ -q
```
- `tests/test_domain.py` β offline, parses the three real captured listing pages (press releases, articles, site search) plus three real document pages (a press-release statement with English translation, an article, a speech transcript)
- `tests/test_live.py` β cautious live requests against www.bundesregierung.de
Tests: 29/29 green (22 offline + 7 live, 2026-11-22).
## License
[PolyForm Noncommercial 1.0.0](https://polyformproject.org/licenses/noncommercial/1.0.0/) β free for personal and non-commercial use; commercial use needs a separate license (see LICENSE).
---
Built by [Mohammad Abbasi](https://github.com/mohammad-abbasi-sh) β [Almanibama](https://almanibama.ir) Β· [mabbasi.ir](https://mabbasi.ir)
TDQS
Scored across 6 tools
Each tool maps to a distinct content type or action: three listing endpoints split by content kind (press releases, articles, speeches), a cross-site search, a document fetcher, and a reachability probe. The closest pair is breg_press_releases vs breg_articles, but the descriptions clarify the difference (teasers vs. articles with description text) and the returned `type` field disambiguates further.
All names share a consistent `breg_` prefix and are snake_case, so the family is immediately recognizable. Minor deviation: the listing tools are bare plural nouns (breg_press_releases, breg_articles, breg_speeches) while the others use verb style (breg_search, breg_get_document, breg_health), but the pattern remains readable and predictable.
Six tools is well-scoped for a read-only news/press site integration: three content listings, one search, one document fetch, and one health check. No tool is redundant and nothing feels missing from the surface that would require another endpoint.
The read-only lifecycle is complete: list per content type, search across all types, and fetch full text (with English translation) via a shared document URL. Gaps are minor β no date-range or category filtering on listings/search and no archive-browsing tool, but agents can work around these by paging and searching.