Junifye MCP Server

Publifye MCP Servers — Scripture research, book authoring, dictionary building and the Norwegian company register for AI clients

Publifye AS runs four hosted Model Context Protocol servers. Point Claude, Cursor, VS Code or any MCP client at them and your assistant can read the Hebrew and Greek text of Scripture, write and typeset a book, build a dictionary, or look up Norwegian organisations in Enhetsregisteret — over an authenticated HTTPS endpoint, with no local install.

This repository is the canonical public reference for those servers: their endpoints, their complete tool schemas, and the provenance of every dataset behind them. It contains no server code — the services are hosted and commercial. What it contains is everything you need to evaluate them before you pay for anything.

Transport is Streamable HTTP. Authentication is OAuth 2.1 with PKCE (S256) and Dynamic Client Registration, or a personal API key — except Brreg, which is OAuth only. See docs/connect.md.

---

Darash — the Hebrew and Greek text, as data

Most Bible tools give a model an English verse. Darash gives it the underlying data: which Hebrew stem a verb is in, which Strong's numbers actually occur in a given verse, what a 19th-century lexicon says about a lemma and which print edition that lexicon came from.

Measured on the live service, 2026-09-13:

Every one of those figures comes from the service's own get_health, not from a brochure. Regenerate them yourself once you have a key.

Why a translation is not enough. A Greek verb in the active voice can be rendered perfectly naturally in the passive in English. An argument about who is acting on whom, built on that English passive, is an argument about the translator. get_morphology returns the voice, person and stem the text actually carries, and strongs_in_verse proves a lemma is really in the verse rather than merely nearby in a concordance.

→ services/darash · tool schemas · PROVENANCE.md — read this one first if you are evaluating the data

Junifye — write the book, then publish it

Chapter-structured authoring over MCP: blocks, headings, Scripture quotations that resolve against a real Bible rather than the model's memory, tables, figures, footnote-free typesetting. Output is a print-ready PDF with the correct binding gutter, screen-reading PDFs, a valid EPUB3 and a web reader. Right-to-left scripts are supported end to end, and editions can be linked as translations of one another.

→ services/junifye · tool schemas

Lexifye — dictionaries with versions and sources

Entries, senses, per-definition sources, revision history and restore. Built for terminology work that has to be defensible later: every definition can carry where it came from, and every change is recoverable.

→ services/lexifye · tool schemas

Brreg — Norwegian organisations, by number or by name

Lookup by organisasjonsnummer, filtered search, and organisational structure over Brønnøysundregistrene's Enhetsregisteret, with key financials from Regnskapsregisteret fetched only when a call asks for them. A name returns candidates, never a silent pick. Open data under NLOD 2.0, attributed in every successful result; not the authoritative register, no roles or persons, no bulk export.

→ services/brreg · tool schemas · PROVENANCE.md · DATA-HANDLING.md

---

Three ways in

You study the biblical languages. Darash is the one that matters. Start with word_study, get_morphology and strongs_in_verse, and read the provenance table before you trust a gloss.

You want it run privately. Darash ships as a single self-contained binary of roughly 205 MB, carrying CLI, HTTP API and MCP server together. It needs no network at all. That is offline Scripture research infrastructure, not a reading app with an export button — and for institutions whose data cannot leave the building it is usually the only workable shape. Licensed by quote.

You are embedding Scripture data in your own product. The complete input schemas, output shapes and tool annotations are in this repository, so you can design against them before you hold a credential. Every Darash tool carries readOnlyHint, destructiveHint, idempotentHint and openWorldHint, derived from its access level rather than hand-written, so a client can plan parallel reads safely.

What this repository does not contain

No server source, no corpus, no lexicon text. Darash's data is roughly 21.5 GiB and the engine is a commercial product; publishing it is a business decision and it has been decided against. What is published here is the description — schemas, provenance, endpoints — which is the part you need to evaluate the service and the part we are willing to be held to.

Publifye AS

Norwegian publishing house, Oslo. Organisation number 826774622 (Enhetsregisteret · Wikidata Q141436507). Company site publifye.com · Privacy · Terms

Licence for this repository: CC BY 4.0. The datasets it describes carry their own licences, recorded per source in each service's PROVENANCE.md.

Written about this work

• Thayer's Greek Lexicon, rebuilt from the 1889 page scans
• Darash vs Logos, Accordance and Blue Letter Bible — an honest comparison
• How to connect Claude, ChatGPT or Cursor — MCP and DCR explained
• Deuteronomy 4:29 says "seek" twice. The Hebrew does not.
• Two arks, one word

Each is published in English, Norwegian, Spanish, Chinese and Korean at blog.publifye.com.