Connect Scriptura to your AI (MCP)

Set it up

Create a developer key

Sign in and go to Profile → Developer keys & MCP. Give the key a name you'll recognize later (“my laptop”), and press Create a key.

The key is shown once — we store only a hash of it, so we genuinely cannot show it to you again. Copy it before you leave the page. If you lose it, revoke it and make another; that's normal and costs nothing.

Paste the connect command

That same screen prints a finished command with your key already in it. Copy it and run it — you don't need to edit anything. For reference, this is the shape:

Claude Code

claude mcp add --transport http scriptura https://YOUR-DOMAIN/mcp \
  --header "Authorization: Bearer sk_dev_your_key_here"

Check it worked: run claude mcp list — Scriptura should say connected.

Claude Desktop & other MCP clients

Add this to your client's MCP config file:

{
  "mcpServers": {
    "scriptura": {
      "type": "http",
      "url": "https://YOUR-DOMAIN/mcp",
      "headers": { "Authorization": "Bearer sk_dev_your_key_here" }
    }
  }
}

Restart the client afterward. Config file locations:

macOS — ~/Library/Application Support/Claude/claude_desktop_config.json
Windows — %APPDATA%\Claude\claude_desktop_config.json

That's the whole setup. Ask your assistant something like “Use Scriptura to show me the morphology of John 1:1 and explain any grammar rules firing there” and it will call the tools directly.

What your AI can do once connected

Tool	What it does
Research tools — start here
word_meaning	Every layer we hold on what a word means, ordered by what kind of evidence each is — published lexicon, the corpus's own glosses, our translation's renderings, computed usage groups, and our word note (paid plans; a free key sees the lock).
word_note	Our own four-part note on a word — In short / Uses / Watch for / Don't over-read. AI-drafted, unverified, never evidence. Part of Starter and up.
sense_groups	A word's occurrences grouped by the contexts they sit in, labelled by the words that distinguish each group. Verse-context groups, not senses — a reading list.
word_study	A whole word study in one call: lexicon entries, exact total occurrences, distribution across books, which inflected forms it takes and how often, and a sample of occurrences from across the canon.
passage_brief	Orientation to a passage: the text, the words most worth examining (least frequent first), the grammar rules firing, and the textual variants — one call instead of four.
interlinear	Reverse interlinear — for each English word of Scriptura Formal, the Greek or Hebrew word behind it, with a confidence and the method on every link. Computed from our own translation, so it is ours to give away; unlinked words are shown, not hidden.
renderings	How a word was actually translated: every English rendering of a source word with counts, or every original behind one English word — where the translation flattened a distinction.
explain_parse	Decode a parse code (V-2AAI-3S, HTd/Ncmpa) into named grammar, or parse every word of a passage. Tables corrected against this corpus, not copied from a textbook.
paradigm	Every form of a word that actually occurs, laid out as a paradigm with the count and an example per cell — and the empty cells left visibly empty.
compound_search	Words in relation — proximity, order and agreement: "an aorist imperative within four words of a genitive noun", "article + adjective agreeing in case, number and gender". Five operators, exact counts.
syntax_search	Search the MACULA syntax trees rather than the string: subjects, objects, predicates, prepositional phrases, genitive chains. Tagged as an interpretation, never as the text.
syntax_conflicts	Where MACULA's analysis and the corpus morphology disagree about a word — substantive disagreements separated from notation. Neither side is corrected to the other.
verb_frames	The valency of a Greek verb from the clause patterns MACULA records, or every verb attested with a given frame — "which verbs take a double accusative", counted.
masoretic_phrasing	How the Masoretic accents divide a Hebrew verse: the accent on every word and the division tree. The Hebrew answer to syntax search.
rule_accuracy	The grammar detectors' own report card: firings, heuristics disclosed, share read off the syntax tree, documented-example recall, and precision only where a human has labelled a sample.
compare_mt_lxx	What the Septuagint put where a Hebrew word stood, or the Hebrew behind a Greek word — statistical alignment, not a translator's note. Non-commercial keys only (LXX-derived, CC BY-NC-SA).
semantic_divergence	Where a Greek word's New Testament usage parts from its Septuagint usage. A prompt to look, not a finding. Non-commercial keys only.
quotation_text_form	When the New Testament quotes the Old: the evidence for whether the wording follows the Greek or the Hebrew, with the diagnostic words that decide it. Non-commercial keys only.
book_profile	What is lexically distinctive about a book: hapax legomena, vocabulary concentrated here rather than merely frequent, and exact totals. “What is characteristic of Jude?”
compare_words	Two to eight words side by side — a pair (ἀγάπη vs φιλία) or a whole word-group. Frequency, distribution, dominant forms, and the books where several of them co-occur, which is where an author draws the distinction himself.
grammar_search	Grammatical concordance: give a parsing pattern (`V-AAM*` = aorist active imperative) and get an exact count, the spread across books, and which precise forms matched.
search_construct	Construct search — the Accordance/Logos power-user query: find a form matching one set of criteria within N words of another (e.g. a present participle within 5 words of θεός in Romans). Each side takes a lemma, Strong's number, or parsing pattern; every hit returns both matched words with parse codes and their distance.
search_institutes	Calvin's Institutes (1559), all four books, with the original Latin beside the English and Calvin's own book.chapter.section references so quotations can be verified.
get_institutes_chapter	A full Institutes chapter in four parallel texts: the 1559 Latin, Beveridge, Allen, and our English.
divine_titles	Which Hebrew title stands behind each “LORD”, “GOD” and “Lord” — Jehovah, Elohim, Adonai, Jah, El, Eloah, Shaddai — and which verses carry two at once where English renders them alike. Reviving Thomas Newberry’s 1890 distinction, computed from the Hebrew rather than transcribed from his page.
newberry_system	Newberry’s full sign system, category by category, with a plain statement of what Scriptura reproduces and what it does not.
list_sources	The provenance registry — every corpus, its exact edition, license, and required attribution. Call this before publishing anything built on these tools.
Your own study notes — opt-in
list_my_studies	Your studies, sermons and papers with titles and note counts. Only on a key created with “read my study notes” ticked.
get_study_notes	Every note in one of your studies, tagged by whether you wrote it, clipped it, or saved it from an answer.
search_my_notes	Search everything you've written across all your studies — “what have I already said about propitiation?”
Direct corpus access
get_passage	A passage with every original-language word tagged — surface form, lemma, Strong's number, parsing code, gloss — plus the English.
search_morphology	Individual occurrences by lemma, Strong's number, or morphology code. Capped at 500 rows; for totals and distribution use word_study or grammar_search.
lookup_lexicon	Public-domain lexicon entries for a Strong's number.
search_text	Keyword search across English, glosses, and lemmas (up to 200 verses).
search_semantic	Meaning-based search — finds conceptually related verses that share no keywords (up to 100).
explain_rules	Named grammar rules firing in a passage, Greek and Hebrew, with why each matters.
explain_variants	The textual-variant apparatus — where the major editions disagree, with witnesses and notes on the famous variants.

Your study notes are opt-in and read-only. A key can only reach them if you ticked “read my study notes” when you created it, and that choice is fixed for the life of the key — it cannot be added later. Without it those tools are not merely refused, they are never offered, so there is nothing for a client to call. Nothing on MCP can ever write to or delete your notes. If a key might end up somewhere shared, leave the box unticked and make a second key for your own machine.

Some corpora carry a non-commercial licence (the Septuagint and everything derived from it). Those tools are served only to non-commercial keys, so what you see listed depends on your plan.

Your private study notes are not on this list, and that is deliberate: the vault-notes tool has no remote handler at all, so there is nothing for a remote client to reach. It exists only on the local server you run yourself.

Limits & billing

MCP is included with a paid Magnify plan — there is nothing extra to buy, but the free tier doesn't include it. Starter includes 150,000 corpus queries a month; Connect (the MCP-only tier, $15/mo) and Pro both include 500,000, which for one researcher is effectively unlimited. We can be that generous because an MCP query is a database read on our side and the AI doing the reasoning is yours — you're spending your own Claude subscription, not ours.

If you're on the free tier you can still create a key and connect; the tools will list, and the first corpus query will tell you to upgrade. Upgrading activates the key you already have — you don't need to make a new one.

Connecting, listing tools, and pinging are free — only actual corpus queries count, and the research-shaped tools cost the least because they return summaries rather than bulk rows. Your current usage and allowance are shown next to each key on the Profile page.

When a key runs out, tool calls return a clear “quota reached” message rather than failing silently, and the connection itself keeps working so you can still see what happened. Allowances reset at the start of each month.

If something goes wrong

What you see	What it means
401 / “requires a key”	No key reached the server. Check the header is exactly `Authorization: Bearer sk_dev_…` — the word `Bearer` and the space matter.
401 / “not valid”	The key was mistyped, truncated on copy, or revoked. Make a new one.
429	Monthly allowance used up. It resets next month, or upgrade your plan.
Tool not found	You asked for a local-only tool (vault notes). Remote servers don't expose it.
Connects, no tools	Restart the client — most read MCP config only at startup.

Attribution

Data reaching you through these tools carries license obligations. If you publish work built on it, credit: original-language text and morphology from STEPBible TAGNT/TAHOT, CC BY 4.0, www.STEPBible.org (Tyndale House, Cambridge). Full details on the Sources & Attribution page.