Bible CLI

A Bible study operating system for AI agents — and the humans working with them.

bible is an offline command-line toolkit that lets any coding agent (Claude Code, Codex, opencode, or anything with a shell) do serious biblical study from the text itself: plain reading, original-language word study, morphology search, cross-translation comparison, and canon-wide pattern analysis — with every claim traceable to a query instead of to training data.

$ bible interlinear "John 3:16" | head -6
John 3:16
Οὕτως     Houtōs    G3779  Adv       so
γὰρ       gar       G1063  Conj      For
ἠγάπησεν  ēgapēsen  G0025  V-AIA-3S  loved
ὁ         ho        G3588  Art-NMS   -
Θεὸς      Theos     G2316  N-NMS     God

$ bible grep-morph --stem niphal --tense participle --book Isaiah --count
98 matching words
  H0539 אָמַן  8
  H5375 נָשָׂא  7
  ...

$ bible freq --strongs H1285 --by-book     # where does "covenant" concentrate?
$ bible xref "Isa 53:5" --text             # how does the canon take this verse up?
$ bible compare "Rom 8:1" -t all           # where do translations diverge?

Everything is offline, deterministic, and open-licensed: two SQLite databases built from public-domain and CC-BY scholarly sources (see Data), downloaded once on first run.

Why

LLMs already "know" the Bible — approximately, unverifiably, and shaped by whatever their training data emphasized. Bible CLI gives an agent the opposite: a set of precise instruments over the actual text, so study becomes evidence-based. The agent reads the passage it cites, checks what a Greek or Hebrew word actually means by how it is actually used, counts patterns instead of gesturing at them, and reports what the text says as distinct from what it concludes. The bundled methodology (bible agent-setup) teaches exactly that discipline — a biblical theology workflow: plain reading → lexical analysis → context → canonical patterns → labeled synthesis.

Related MCP server: Bible MCP

Install

npm install -g bible-cli        # or: npx bible-cli ...
bible db download               # fetches the databases (~100 MB once)
bible passage "John 3:16"

For agents, add the study methodology to your project:

bible agent-setup claude       # writes .claude/skills/bible-study/SKILL.md
bible agent-setup generic      # appends a section to AGENTS.md (codex, opencode, …)

Or run it as an MCP server: bible mcp (stdio) exposes all commands as tools.

Commands

Every command supports --json (machine-readable output), forgiving references (John 3:16-18, jn 3 16, 1jn2:5, Psalm 23, Gen 1:1-2:3), scoping (--book Isaiah, --book Gen-Deu, --book ot), and helpful errors (Unknown book 'Pslams'. Did you mean: Psalms…).

The data

All redistributable, all attributed (see bible licenses):

† The LXX ships as a separate optional artifact (bible db download-lxx) because its digitization carries CC BY-SA; the core and study databases stay public domain + CC BY only.

The databases are built by a reproducible pipeline in this repo (npm run pipeline) with a verification stage that checks canonical verse counts, versification edge cases (Psalm titles, Malachi 3/4, Joel 2/3), Ketiv/Qere handling, and morphology-code coverage — the build fails loudly rather than shipping silently wrong data.

Design notes

• Verse IDs are integers (BBCCCVVV; Gen 1:1 = 1001001) on a KJV-English spine; Hebrew/Greek versification differences are mapped at build time and preserved for display. Psalm superscriptions are verse 0.

• One row per morpheme: Hebrew prefixes/suffixes and Greek crasis components are individually tagged, so grep-morph and frequency counts are exact. Analytics default to the default text stream (Qere; NA-stream Greek) so textual variants never inflate counts.

• The CLI is the single source of truth; the MCP server shells into it, so both interfaces always agree.

• Quotation detection is computed, not curated: bible quotes reports runs of 5+ identical normalized Greek words shared between the NT and the LXX. Run length is evidence strength. It measures verbal quotation; thematic allusion without shared wording is bible xref's domain.

Known limitations (honest edges)

• --edition Greek texts are reconstructed from TAGNT's word-set + variant apparatus — accurate wording, but not a facsimile of a printed edition's punctuation or orthography.

• ~80 Aramaic Strong's numbers lack formal lexicon entries (gaps in the BDB mapping); their entries are synthesized from the tagged text's own glosses and labeled as such.

• English reverse lookup (bible word lovingkindness) is heuristic: lexicon glosses first, then which original words underlie verses containing the English word.

• The Swete LXX digitization lacks Ecclesiastes; Daniel uses Theodotion (which the NT normally follows).

Roadmap

Semantic similarity via local embeddings, pericope/discourse boundaries, versification traditions beyond Hebrew/Greek (Vulgate), additional open translations (non-English included), and syntax-aware search (MACULA trees). PRs welcome.

License

Code: MIT. Data: see docs/DATA-SOURCES.md and bible licenses — public domain, CC BY 4.0, and CC0 in the core/study databases; the optional LXX artifact is CC BY-SA 4.0.