# Scythian dictionary access

> A grounding service over the English Wiktionary's Proto-Scythian entries — the
> reconstructed Eastern Iranian language of the steppe nomads (Scythians,
> Sarmatians, Saka), ancestor of Ossetian and Khotanese. The corpus is a few
> dozen reconstructed lemmas (asterisked: *tiɣri 'bitter'); use it to cite real
> reconstructions and their descendants instead of inventing "Scythian" words.
> Read-only.

## What it does
- Looks a Proto-Scythian lemma up (any spelling: plain ASCII is folded — tigri
  finds *tiɣri, θ may be typed t or th, ɣ→g, δ→d, β→b, ə→e) and returns its
  senses plus the full descendants payload: etymology, English glosses and the
  reflex tree through Sarmatian→Alanic→Ossetian and Saka→Khotanese.
- Finds the lemmas matching an ENGLISH word (search_language=eng): the query is
  matched against every entry's indexed glosses (reconstructed languages have no
  Wiktionary translation tables), then expanded into full entries. Most English
  words match nothing — the corpus is tiny; try simple concrete words ('horse',
  'gold', 'bitter').
- Proto-Scythian is reconstructed: it has NO inflection tables, so there is no
  inflections tool here — get_descendants_scythian replaces it, and every form
  carries the scholarly asterisk.
- A one-time corpus sweep on the server's first start builds the complete
  lemma + gloss index (an eng query during that minute asks you to retry); a
  lemma added upstream is picked up by the next restart's sweep.

## How to connect (MCP)
Remote streamable-HTTP endpoint (no auth, read-only): https://scythian.valksor.com/mcp
  Claude Code:  claude mcp add --transport http scythian https://scythian.valksor.com/mcp
  Cursor / Cline (~/.cursor/mcp.json):
    {"mcpServers": {"scythian": {"url": "https://scythian.valksor.com/mcp"}}}
  stdio-only clients — bridge with mcp-remote:
    {"mcpServers": {"scythian": {"command": "npx",
       "args": ["-y", "mcp-remote", "https://scythian.valksor.com/mcp"]}}}
Tools (each returns Markdown plus typed structuredContent per its outputSchema):
- search_scythian(query, search_language='xsc'|'eng', include_forms=true, max_forms=N)
    Senses plus descendant handles (entry_id = the bare lemma, word_class = the
    part-of-speech section). With include_forms (default TRUE on MCP; the JSON
    API defaults it false), each match's full descendants payload comes back
    inline. With search_language='eng' the query is an English word and the
    result lists the gloss-matched lemmas plus their entries (English→Scythian).
- get_descendants_scythian(entry_id, word_class)
    The full descendants payload for one handle. Its structuredContent nests the
    payload under a "result" key (the search tool's payload is unwrapped); the
    JSON API returns it directly. Both shapes are discriminated by "category":
    'descendants' gives headword, etymology, glosses and tree (nested
    {language, form, transliteration, qualifier, children} nodes — form keeps
    the asterisk on reconstructions and may be empty for branch headers);
    'not_found' when no entry exists.
Search tips: one word per call; the reconstruction asterisk is optional; for
English→Scythian pass search_language='eng' with one simple content word.

## How to call (JSON API)
- GET https://scythian.valksor.com/api/search?query=tigri
- GET https://scythian.valksor.com/api/search?query=bitter&search_language=eng
- GET https://scythian.valksor.com/api/descendants?entry_id=tiɣri&word_class=adjective
- GET https://scythian.valksor.com/api/health
- Interactive docs: https://scythian.valksor.com/api/docs  ·  schema: https://scythian.valksor.com/api/openapi.json

## Caching and politeness
Dictionary responses are cached server-side with a long TTL (Wiktionary content
changes slowly). Upstream calls are serialised with a polite minimum interval.
Calls are rate limited per client IP — on a 429, wait a few seconds and retry.

## Source and licence
- Dictionary: Wiktionary, the free dictionary (en.wiktionary.org), Proto-Scythian reconstruction entries. Wikimedia Foundation. Text available under CC BY-SA 4.0; attribution and share-alike required when republishing.
- Licence: CC BY-SA 4.0 — attribute and share alike when republishing dictionary text.