kansallisarkisto-mcp

Connect

Hosted on Hugging Face over streamable HTTP — no key, no sign-up:

https://riksarkivet-kansallisarkisto-mcp.hf.space/mcp

Claude Code

claude mcp add --transport http kansallisarkisto \
  https://riksarkivet-kansallisarkisto-mcp.hf.space/mcp

claude.ai

Settings → Connectors → Add custom connector, with the URL above.

The Space sleeps after 48 hours without traffic. The first request wakes it, so a client that times out once will usually connect on the second try.

Run it yourself (stdio, from a clone)

make install
make harvest            # df only: 6,876 charters, ~3 s
make ingest-df          # builds data/df from .data/df/df.jsonl.gz
claude mcp add kansallisarkisto -- uv run kansallisarkisto-mcp

Docker (streamable HTTP on :8000)

docker run -p 8000:8000 \
  -v "$PWD/data:/data:ro" \
  riksarkivet/kansallisarkisto-mcp:latest

The image ships without data: mount a LanceDB directory at /data, or point KA_LANCEDB_URI elsewhere. Without it the server still boots and the tools say so.

MCP Inspector

make inspect

Tools

• df_search — Full-text search over Diplomatarium Fennicum: 6,876 medieval charters concerning Finland, 859–1530. Filter by language, place and country of issue, and year range; page with offset.
• df_get_charter — One charter's full transcript and catalogue record by its DF number, the citable identifier carried on every hit.
• voudintilit_search — Full-text search over voudintilit: 98,945 pages of the Häme and Satakunta bailiffs' accounts, 1539–1635. Filter by bailiwick, account book and year; every hit cites its reference, account book, year and page, and links to the page image in Astia.
• voudintilit_get_page — One page's full text by its page id, with the previous and next pages of its volume.
• tuomiokirjat_search — Full-text search over tuomiokirjat: 7.8 million pages of Finnish lower-court records, 1610–1931, from 223 archives. Filter by archive, series and year; every hit cites its series, signum, year and page, and links to the page image in Astia.
• tuomiokirjat_get_page — One court-record page's full text by its page id, with the previous and next pages of its volume.

Search tips

• The text is not in Finnish. Finland was part of the Swedish realm until 1809, so these documents are early-modern Swedish, Latin and German. Only the catalogue metadata is Finnish.
• Search in period spelling: bref not brev, konung not kung, Åbo not Turku, Viborg not Viipuri.
• Swedish stemming and accent folding are applied — konungen finds konung, Abo finds Åbo.
• Several words must all appear; "quoted words" are an exact phrase; bref|breff matches either spelling. On the charters a trailing * is a prefix — lepros* reaches the Latin and German the stemmer does not — while the two page corpora are too large for one, so list spellings or use fuzzy=1.
• Language labels are unaccented Finnish: venaja, never venäjä.
• 36% of the charters are catalogued but never transcribed. They are still returned — findable by place, index term and language — and marked as untranscribed.
• Voudintilit and tuomiokirjat are paged: each hit is one page of a volume, cited by its archival reference and linked to its image in Astia, and the page tools name the previous and next pages — accounts and cases run across pages. On the 7.8 million court pages common words match more than 10,000; narrow by archive, series or years.
• Transcripts are OCR/HTR output over a millennium of handwriting. Check any quotation against the archive.

Learn more

Parameter reference, corpus notes, and how the server is built and deployed:

GitHub · Sisältöhaku · Kansallisarkisto