圣经学习MCP服务器

Study Bible MCP Server

A Bible study assistant for Claude that provides full scholarly lexicons (LSJ Greek, BDB Hebrew, Abbott-Smith NT Greek), morphologically-tagged biblical texts, cross-references, Theographic genealogy graphs, Aquifer Open Study Notes, a Bible dictionary, key theological terms, Ancient Near East cultural context, and hermeneutical methodology based on Fee & Stuart's "How to Read the Bible for All Its Worth".

Quick Start

Claude Desktop

1. Open Settings → Connectors
2. Click Add Custom Connector
3. Paste: https://studybible-mcp.fly.dev/sse
4. Restart Claude Desktop

That's it. No signup, no API keys, no downloads.

Claude Code

claude mcp add study-bible https://studybible-mcp.fly.dev/sse

ChatGPT (Developer Mode)

1. Open Settings → Developer Mode → enable it
2. Go to Actions → Add MCP Server
3. Enter URL: https://studybible-mcp.fly.dev/mcp/

Cursor / Windsurf / Cline

Add to your MCP configuration (usually .cursor/mcp.json, .windsurf/mcp.json, or Cline settings):

{
  "mcpServers": {
    "study-bible": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://studybible-mcp.fly.dev/sse"]
    }
  }
}

Install via pip / uvx (local stdio)

pip install studybible-mcp
studybible-mcp --transport stdio

Or run directly without installing:

uvx studybible-mcp --transport stdio

Manual Config (Alternative)

If you prefer editing config files directly:

{
  "mcpServers": {
    "study-bible": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://studybible-mcp.fly.dev/sse"]
    }
  }
}

Config file locations:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json
• Linux: ~/.config/Claude/claude_desktop_config.json

How It Works

Architecture Overview

┌─────────────────────┐     ┌──────────────────────────────────────────┐
│   Claude Desktop    │     │   Study Bible MCP Server                 │
│   or Claude Code    │     │   (Fly.io)                               │
│                     │     │                                          │
│  ┌───────────────┐  │ MCP │  ┌─────────────┐  ┌───────────────────┐  │
│  │ User asks a   │──┼─────┼─▶│ MCP Server  │─▶│ SQLite DB (600+MB)│  │
│  │ Bible question│  │     │  │ (Python)    │  │                   │  │
│  └───────────────┘  │     │  └─────────────┘  │ • Lexicons (LSJ,  │  │
│                     │     │        │          │   BDB, Abbott-     │  │
│  ┌───────────────┐  │     │        ▼          │   Smith, Strong's) │  │
│  │ Claude uses   │◀─┼─────┼── Tool Results    │ • Tagged NT + OT  │  │
│  │ study tools to│  │     │                   │ • Names + ACAI    │  │
│  │ answer        │  │     │                   │ • Morphology      │  │
│  └───────────────┘  │     │                   │ • Study Notes     │  │
│                     │     │                   │ • Bible Dictionary │  │
│                     │     │                   │ • Key Terms        │  │
│                     │     │                   │ • ANE Context      │  │
│                     │     │                   │ • Embeddings       │  │
│                     │     │                   └───────────────────┘  │
└─────────────────────┘     └──────────────────────────────────────────┘

When you ask Claude a Bible question:

1. Claude analyses your question and decides which tools to use
2. Claude calls the appropriate MCP tools (e.g., lookup_verse, word_study, get_study_notes)
3. The server queries the pre-built SQLite database
4. Results are returned to Claude
5. Claude synthesises the data into a helpful response

See ARCHITECTURE.md for a full Mermaid flowchart of the available tools and how the agent chains them together.

The Database

The server includes a pre-built SQLite database (~600MB) containing:

You can download the pre-built database directly: study_bible.db (~750MB)

The database includes CC BY-NC 4.0 material (the ETCBC Dead Sea Scrolls transcription and the Metsudah Chumash English), for non-commercial use only, and CC BY-SA 4.0 material (Perseus Josephus, First1KGreek Philo, Latin Wikisource 4 Ezra), which must be shared alike. The download response repeats this in an X-Data-License-Notice header.

Available Tools

Core Text & Language

1. lookup_verse

Retrieve a verse, verse range, or whole chapter with its original language text and word-by-word analysis.

When Claude uses it: User asks about a specific verse or passage, wants to see the Greek or Hebrew text, or asks about specific words in a passage.

Accepted references: a verse (John 3:16), a range within a chapter (Luke 19:1-3; an en dash also works), a cross-chapter range (Luke 19:45-20:2), or whole chapters (Psalm 23, Psalm 23-24). Single-chapter books (Obadiah, Philemon, 2 John, 3 John, Jude) may be cited by verse alone, as is conventional: Jude 3 means Jude 1:3 and a bare Jude means the whole book. This applies to every tool's reference argument. Verses are matched by English chapter/verse numbering, so Old Testament verses whose Hebrew numbering differs (e.g. Psalm 40:6, stored as Psa.40.6(40.7)) are found, and an English verse stored as several source rows returns all of them. A Psalm superscription, which is verse 1 in the Hebrew, is headed as such (Psa.82 superscription (Hebrew Psa.82.1)). A passage returns at most 50 verses; longer requests (e.g. Psalm 119) open with a note naming the reference that continues the passage. Unknown books, missing chapters or verses, and reversed ranges get a message saying what exists.

What it returns: English text from the Berean Standard Bible (labelled (BSB) once per response; STEPBible's word-by-word gloss, marked [gloss], stands in for the few verses the BSB lacks, such as Psalm superscriptions and Matthew 17:21), original Greek/Hebrew text, word-by-word breakdown with Strong's numbers, optional morphological parsing, and genre-specific interpretation guidance. Notes about cross-references, Torah Weave partners, and NT/OT LXX quotations appear once per lookup; for a passage they name a verse to pass to those single-verse tools.

Example:

User: "Look up Romans 8:28 with the Greek"

Claude calls: lookup_verse(reference="Romans 8:28", include_original=true)

Returns:
- Verse text in English
- Greek: πάντα συνεργεῖ εἰς ἀγαθόν...
- Word analysis: συνεργεῖ (G4903) = "works together" [V-PAI-3S]
- Genre guidance: "Epistles - What problem was being addressed?"

compare_biblical_versions

Compare the Masoretic Text and Septuagint with explicit per-verse provenance. Covers Old Testament and Septuagint books only. Rows that share a canonical (MT/English) reference appear under one header, while each row keeps its own edition reference, so Hebrew/Greek Psalm numbering and other mapped versification shifts stay visible. Psalm 82:6, for example, shows MT Psa.82.6 and LXX Psa.81.6; Psalm 40:6 shows MT Psa.40.7 (Hebrew numbering) and LXX Psa.39.7.

compare_biblical_versions(
  reference="Psalm 24:1",        # also "Psalm 82:1-8", "Psalm 81:6-82:1", "Psalm 82"
  versions=["MT", "LXX"],
  numbering="canonical",          # or "lxx" to read the reference in Greek numbering
  include_original=true,
  include_morphology=false
)

Each edition row names its version, edition, and edition reference (plus the versification mapping whenever it is not identity), then its original text and English. Licences, source URLs, translation labels, the verse-mapping credit, and the Brenton caveat are listed once in a closing Sources block rather than on every row. A call returns at most 10 verses; a longer range or chapter is cut there with a note naming the verse to continue from.

• MT comes from the STEPBible TAHOT verses already in the database (Westminster Leningrad Codex). Its English is STEPBible word glosses (interlinear, not a translation).
• LXX covers all 59 books of the Rahlfs source edition. Brenton's English is shown only for the recension Brenton translated (Theodotion for Daniel, Susanna and Bel; Vaticanus B for Joshua and Judges); Old Greek and Alexandrinus rows say the translation is not available for that edition.
• Brenton's English is not a translation of the Rahlfs text. Brenton (1844) translated his own Greek edition, based chiefly on Codex Vaticanus, and his English sometimes departs even from that Greek. Every row with his English carries a note saying so. Each such verse is also compared with Brenton's own Greek after spelling noise is removed: accents, breathings, case, punctuation, elision, movable nu, itacism (ει/ι), doubled consonants, nasal assimilation, crasis, Hellenistic spellings (λήμψομαι/λήψομαι), word division (ἀνὰ μέσον/ἀναμέσον), name spellings, and one-letter keying errors in Brenton's digital text. Where his Greek still adds, lacks, substitutes, or reorders words, the row shows a warning, a difference summary, and Brenton's Greek as the text his English translates. Deuteronomy 32:43, for example, reports "word order differs (υἱοὶ ἄγγελοι)", and Psalm 40:6 reports "Brenton's Greek reads σῶμα for ὠτία". Differing verse divisions do not count as differences. 8,914 of the 23,684 rows with Brenton's own English (38%) are flagged. Unflagged rows can still differ, because a Greek-to-Greek comparison cannot see where his English departs from his own Greek.
• Most Apocrypha English is the KJV, not Brenton's translation. For 1 Esdras, 1–2 Maccabees, Baruch, Bel, Judith, the Letter of Jeremiah, Sirach, Susanna, Tobit, Wisdom, and Daniel 3:24–90 (the Prayer of Azariah and Song of the Three), Brenton's edition prints the KJV (Authorized Version) Apocrypha, lightly adapted by Brenton (verse-by-verse similarity to the KJV is 92–100%, against 24–49% for his own Hebrew-canon translations). The KJV was translated from other Greek texts, so it can contain words neither the Rahlfs nor Brenton's Greek has: Sirach 17:17 adds "in the division of the nations of the whole earth". These 4,630 rows label the English "KJV Apocrypha", are named in the Sources block with one closing note, and are not compared with Brenton's Greek. Greek Esther and 3–4 Maccabees are Brenton's own translation.
• Apocrypha rows are marked. Rows outside the 66-book Protestant canon (the Septuagint-only books, the Greek additions to Esther, Daniel 3:24–90, Psalm 151, and Odes 7–8, 12 and 14) end their header with "(Apocrypha)", and the closing block adds one line saying they are not in the Protestant canon (cf. Westminster Confession 1.3). Other Septuagint pluses in canonical books (3 Kingdoms 2:35a–o, Proverbs 9:12a–c, ...) are textual variants and are not marked.
• LXX-only verses (pluses such as the Prayer of Azariah, Psalm 151, and the Septuagint-only books) have no canonical MT position, so they never appear as the counterpart of a different MT verse. Request them with numbering="lxx"; a canonical lookup notes when such verses sit at the same Greek-numbered position. Septuagint-only books (Tobit, Sirach, ...) always use Greek numbering.

New Testament references, unparseable references, and references with no matching verses each return a specific message. A database without the Septuagint table tells you to run python scripts/upgrade_database.py.

search_ancient_sources

Search licensed primary-source corpora for Jewish history. corpus is any registered corpus key or all; the registry (src/study_bible_mcp/corpora/) currently holds josephus: 2,304 Whiston sections from Jewish Antiquities, The Jewish War, Life, and Against Apion. Each section is paired with every Niese Greek section it spans and referenced as a Niese span plus Whiston's numbering, e.g. Antiquities 15.317–322 (Whiston 15.9.3). targum holds Targum Onkelos on the five books of the Torah: 5,846 Aramaic verse passages with the CC BY-NC Metsudah English, an English of the Hebrew in which Onkelos's own renderings are marked ⟨…⟩ (see Targums), each linked to the verse it translates. qumran holds 44 Hebrew manuscript columns with numbered lines, no English translation included: the Damascus Document (CD 1:1–21, Genizah MS A columns 1–16 and MS B columns 19–20), the Community Rule (1QS), the Rule of the Congregation (1QSa) and Pesher Habakkuk (1QpHab), all from the ETCBC/dss transcription; 32 curated links name the Habakkuk verses each 1QpHab lemma quotes. philo holds the 31 Greek treatises of Philo of Alexandria (Cohn–Wendland, via First1KGreek), one passage per Cohn–Wendland section with the same section of Yonge's English (7,901 passages; 58 sections of Spec. 1–2 that Yonge's text lacks are Greek only). pseudepigrapha holds 1 Enoch in R. H. Charles's English, one passage per Charles verse, and 4 Ezra (2 Esdras 3–14) in the KJV with the Clementine Latin, one passage per verse (1,679 passages); both works are non-canonical and are labelled as attributed to Enoch or Ezra. The rabbinic corpus holds the Mishnah from Sefaria (63 tractates, one passage per mishnah). Each result gives its citation and, on the line below, its canon-and-date tag (e.g. *Not Scripture · Jewish pseudepigraphon (Parables, c. 50 BC–AD 117, debated) · tier A/B · Charles 1912*), then up to 1,500 characters of English around the first match and a shorter original-language excerpt (match-centred when the query is in the original's script; a passage without English shows its original instead), any editorial ⚠ notes, and its typed biblical links. A closing Caveats block prints each corpus's caveat and the date tiers present once, and a Sources block lists the edition, translation, license (non-commercial licences marked and linked), and pinned source once. Up to 4 passages are returned by default, within a 12,000-character budget (see Ancient sources).

search_ancient_sources(
  query="John Baptist",
  corpus="josephus",
  biblical_reference="Mark 6:20",
  period="second_temple",
  limit=4
)

Only the passage text is searched, so a word that appears only in a work title (Jewish, War, Life) does not match every passage of that work. English search is stemmed (build matches building); Greek, Hebrew, and Aramaic search ignores accents, breathings, vowel points, cantillation, and case (βαπτισμω matches βαπτισμῷ), and editorial signs printed inside a word (יעבורו matches the corrected י{{א}}<<ע>>בורו); Qumran text in [ ] is a modern reconstruction and is never matched. Passages containing every query word come first; when none do, passages matching any word are returned with an explicit note. Only when the exact search finds nothing is a Hebrew or Aramaic word retried with one or two attached prefix letters (ו ב ה ל מ כ ש) stripped from the indexed words, keeping at least three letters (הון then finds ובהון); those results are marked (prefix match), because a root letter can look like a prefix (קדש finds מקדש). biblical_reference accepts a chapter, verse, or range (Luke 21, Luke 21:5, Luke 21:5-6, Jude 6) and filters to editorial links: curated links (33 for Josephus, 2 for the rabbinic corpus), each checked against the linked passage text, links from corpora/editorial_notes.json, and one "Translation of" link per targum verse, derived from its verse key (biblical_reference="Deuteronomy 32:8" with corpus="targum" finds Onkelos on that verse). Each link is typed — Aligned, Echo, Shared tradition, Contrast, or Translation of — and printed as e.g. - Shared tradition: Mark 6:17–29 — Related account: death of John the Baptist. Period tags, biblical links, and editorial chapter headings (Whiston's, for Josephus) are labelled editorial metadata; they are not presented as claims made by the ancient author. Empty results say whether the corpus is missing, the reference is unsupported, the reference has no curated link, or nothing matched.

lookup_ancient_source

Read an ancient-source passage by its citation. A citation is a work name or abbreviation plus a locator in that work's numbering; an author prefix is optional. For Josephus: Antiquities/Ant./A.J., War/B.J./J.W., and Against Apion/C. Ap. take Niese book.section; Life/Vita takes a section. For Targum Onkelos, Targum Onkelos/Onkelos/Tg. Onq. precedes a Torah book (Genesis/Gen, Exodus/Exod, Leviticus/Lev, Numbers/Num, Deuteronomy/Deut) and chapter:verse; the prefix is optional, so a bare Genesis 1:1 also finds Onkelos. Onkelos is cited by English verse numbers, like the rest of the server; where Sefaria's Hebrew (MT) numbering differs, the heading and link show it: Targum Onkelos Exodus 8:5 (MT 8:1), and the Peniel verse is Onkelos Genesis 32:30 (MT 32:31). Sefaria joins the four short commandments into one verse, served as Exodus 20:13–16 and Deuteronomy 5:17–20. For Qumran: CD/Damascus Document/Zadokite Fragments, 1QS/ Community Rule, 1QSa/1Q28a/Rule of the Congregation, and 1QpHab/Pesher Habakkuk take column:line, with Arabic or Roman column numbers (CD 6:11, CD VI 11, 1QpHab VII 1-5); a line citation shows the cited lines ±2 from the column that contains them, and a column citation (CD 6) the whole column, with the line numbers in the text. For Philo, SBL short titles, Latin titles, and English titles (Opif./De Opificio Mundi/On the Creation of the World, Legat./Leg. Gai., Her./Heres, ...) take a Cohn–Wendland section; the multi-book works Leg. 1–3, Somn. 1–2, Mos. 1–2, and Spec. 1–4 take book.section; a book number the work lacks (Leg. 31, likewise Ant. 21.1) is rejected rather than read as a book. 1 Enoch/1 En. and 4 Ezra/2 Esdras/4 Esdras take chapter:verse. 4 Ezra 3–14 is 2 Esdras 3–14 with the same chapter numbers, so 2 Esdras 11:1 and 4 Ezra 11:1 are the same passage; 2 Esdras 1–2 and 15–16 (5 and 6 Ezra) are not included. 4 Ezra is cited in modern (Bensly) numbering: the KJV and Clementine Latin lack the "missing fragment" 7:36–105, so their 7:36–70 is 4 Ezra 7:106–140, shown as e.g. 4 Ezra 7:106 (KJV 2 Esdras 7:36). A 4 Ezra citation outside the served text says what is missing and why. For the Mishnah, a tractate takes chapter:mishnah, with an optional Mishnah/m. prefix (Mishnah Shekalim 1:1, m. Shek. 1:1). Tractate names follow Sefaria (Menachot, Pesachim) with SBL abbreviations and common spellings as aliases (Menahot, Men., Pesah., B. Bat.); apostrophes are ignored, so Ta'anit 2:1 is Taanit 2:1.

lookup_ancient_source(citation="Ant. 15.320")            # inside Antiquities 15.317–322
lookup_ancient_source(citation="Josephus, War 6.316")
lookup_ancient_source(citation="Antiquities 18.116-121", include_original=False)
lookup_ancient_source(citation="Tg. Onq. Deut 32:8")
lookup_ancient_source(citation="Onkelos Genesis 1:1-3")
lookup_ancient_source(citation="CD 6:11")               # inside CD 6:1–21 (CD-A)
lookup_ancient_source(citation="1QpHab 7:4")
lookup_ancient_source(citation="Philo, Legat. 299")
lookup_ancient_source(citation="Leg. 1.31")
lookup_ancient_source(citation="1 Enoch 89:73")
lookup_ancient_source(citation="2 Esdras 11:1")          # = 4 Ezra 11:1
lookup_ancient_source(citation="m. Shek. 1:3")

A locator inside a longer passage returns the passage containing it; a range returns the overlapping passages in citation order, 4 by default (limit up to 10), with a note when more exist. Each passage shows its citation, canon-and-date tag, up to 1,500 characters of English and 600 of the original (full_text=true lifts these caps), editorial ⚠ notes, and typed biblical links; caveats and sources are printed once at the end, and the response stops at about 12,000 characters, naming the citation to continue from. If registered works share a title, a collection prefix (such as Mishnah) or the locator grammar itself selects one. Unknown works get a short summary of the citable works (and a close match, e.g. Did you mean Antiquities?), ambiguous citations list the candidates, unreadable or missing locators name the expected grammar with an example from the same work, a locator outside the imported text names the imported range, the nearest passages, and any gap note covering it, and a database without the corpus (or with ancient-source tables from before the current schema) is told to run python scripts/upgrade_database.py.

2. word_study

Deep dive into a Greek or Hebrew word's meaning, usage, and etymology. Greek entries include both the full Liddell-Scott-Jones (LSJ) definition for classical breadth and the Abbott-Smith definition for NT-specific insight (with LXX/Hebrew equivalents, synonym discussions, and NT occurrence counts). Hebrew entries include the full Brown-Driver-Briggs (BDB) definition.

When Claude uses it: User asks about the meaning of a Greek/Hebrew word, wants to understand different translations of a term, or asks about theological concepts (love, faith, grace).

What it returns: Original word in Greek/Hebrew script, transliteration, Strong's number, brief definition, full LSJ or BDB scholarly definition, Abbott-Smith NT-focused definition (Greek words), LXX/Hebrew cross-references, synonym discussions, NT occurrence count, etymology, usage count, semantic range, related words, example passages.

Example:

User: "What does 'agape' mean in Greek?"

Claude calls: word_study(strongs="G26")

Returns:
- ἀγάπη (agapē, G26)
- Brief Definition: "love, goodwill, benevolence"
- Full LSJ Definition: [scholarly entry with classical and biblical usage]
- Abbott-Smith Definition: [NT-focused entry with sense hierarchy]
- LXX / Hebrew Equivalents: H0160 אַהֲבָה
- Synonyms: SYN.: φιλία — love which chooses its object...
- NT Usage: Occurs 116 times in the New Testament
- Related words: ἀγαπάω (to love), ἀγαπητός (beloved)

3. search_lexicon

Search across Greek and Hebrew lexicons by English meaning.

When Claude uses it: User wants to find the Greek/Hebrew word for an English concept, asks "How do you say X in Greek?", or is exploring related words.

Example:

User: "What are the Greek words for 'love'?"

Claude calls: search_lexicon(query="love", language="greek")

Returns:
- G26 ἀγάπη - divine, unconditional love
- G5368 φιλέω - brotherly, affectionate love
- G2309 θέλω - desire, wish

4. get_cross_references

Find passages related to a verse or theological theme.

When Claude uses it: User asks "What other passages talk about X?", wants to see Scripture interpreting Scripture, or is studying a theological theme.

Built-in themes: salvation_by_grace, deity_of_christ, atonement, resurrection, holy_spirit, justification.

Example:

User: "What passages support salvation by grace alone?"

Claude calls: get_cross_references(theme="salvation_by_grace")

Returns:
- Genesis 15:6 - Abraham believed, credited as righteousness
- Romans 3:21-26 - Righteousness through faith in Christ
- Ephesians 2:8-9 - Saved by grace through faith
- Titus 3:5-7 - Not by works of righteousness

5. lookup_name

Get information about biblical people, places, and things, enriched with ACAI entity annotations (variant names, roles, reference counts, speech attributions).

When Claude uses it: User asks "Who was X?", wants to know about a place's significance, or asks about relationships between biblical figures.

What it returns: Original Hebrew/Greek form, type (person, place, thing), description, key references, relationships (father, mother, children), ACAI annotations (variant names, roles, verse reference count, attributed speeches).

Example:

User: "Who was Abraham?"

Claude calls: lookup_name(name="Abraham")

Returns:
- Type: Person
- Original: אַבְרָהָם
- Description, key references, family relationships
- ACAI: Also known as "Abram", referenced in 275 verses, 52 attributed speeches

6. parse_morphology

Explain Greek or Hebrew grammatical parsing codes.

When Claude uses it: User asks what a morphology code means or wants to understand verb tense, mood, voice.

Example:

User: "What does V-AAI-3S mean?"

Claude calls: parse_morphology(code="V-AAI-3S", language="greek")

Returns:
- Part of Speech: Verb
- Tense: Aorist, Voice: Active, Mood: Indicative
- Person: 3rd, Number: Singular

7. search_by_strongs

Find all verses containing a specific Strong's number.

When Claude uses it: User wants to see how a word is used throughout Scripture or is doing a comprehensive word study.

Example:

User: "Show me all the places where 'hesed' (H2617) appears"

Claude calls: search_by_strongs(strongs="H2617", limit=20)

Returns:
- H2617 חֶסֶד (hesed) - "lovingkindness, steadfast love"
- Genesis 24:12, Exodus 34:6, Psalm 23:6, Psalm 136...

Scholarly Commentary

8. get_study_notes

Get scholarly study notes and translation notes for a Bible verse or chapter.

When Claude uses it: User wants commentary on a specific verse, help explaining a difficult passage, or translation and cultural background notes.

What it returns: Combined commentary from three sources: Aquifer Open Study Notes (66 books of concise verse-level scholarly commentary), unfoldingWord Translation Notes (translator-focused linguistic insights), and SIL Translator Notes (additional translation and cultural context).

Example:

User: "What do the study notes say about John 3:16?"

Claude calls: get_study_notes(reference="John 3:16")

Returns:
- Aquifer Open Study Notes: [verse commentary]
- Translation Notes (UW): [linguistic insights]
- Translation Notes (SIL): [cultural context]

9. get_bible_dictionary

Look up a topic in the Tyndale Bible Dictionary.

When Claude uses it: User needs background information on a biblical topic, historical or cultural context, or a detailed article about a person, place, or concept.

What it returns: The dictionary article from a collection of 500+ topical articles covering biblical people, places, theological concepts, cultural practices, historical background, and archaeological findings. A long article comes about 4,000 characters at a time, ending with the offset to continue from and the headings of the sections still to come (section="The Sinai Covenant" reads one section).

Example:

User: "Tell me about the Pharisees"

Claude calls: get_bible_dictionary(topic="Pharisees")

Returns: Full Tyndale Bible Dictionary article with historical background,
beliefs, practices, and NT context.

10. get_key_terms

Look up a key theological term in the FIA Key Terms database.

When Claude uses it: User wants a precise definition of a theological term, or needs to understand how a concept is used across Scripture.

What it returns: Definition from a collection of 200+ carefully defined theological and biblical terms with biblical usage, cross-references, and translation guidance.

Example:

User: "Define the term atonement"

Claude calls: get_key_terms(term="atonement")

Returns: FIA Key Terms definition with biblical usage, cross-references,
and translation guidance.

Graph Knowledge (Theographic Bible Metadata)

11. explore_genealogy

Trace family relationships (ancestors or descendants, 1 to 40 generations) for a biblical person using the parent links of 2,000 persons.

What it returns: Each ancestor or descendant once, generation by generation, with whose parent or child they are (and curated notes such as Joseph as Jesus' legal father), immediate family (spouse, siblings), and a Mermaid diagram when the tree has at most 40 people. Dictionary articles are left to get_bible_dictionary.

Example:

User: "Show me the family tree of David"

Claude calls: explore_genealogy(person="David", direction="both", generations=5)

Returns: Ancestor chain (Jesse → Obed → Boaz...), descendants,
spouse(s), siblings, Mermaid flowchart diagram.

12. explore_person_events

Find all events a biblical person participated in, in chronological order.

What it returns: Each event once, in chronological order, with its approximate date (Theographic chronology), places, and passage. Names resolve as Scripture uses them ('Peter', 'Jacob' the patriarch), and the tool says which namesake it chose.

13. explore_place

Explore the biblical history of a geographic location.

What it returns: Events at the location in chronological order with their passages, people born or who died there, coordinates, Mermaid network diagram.

14. find_connection

Find the family relationship path between two biblical people.

What it returns: The closest relationship path through parent, child, sibling, and partner relationships, preferring lines of descent, summarised and listed compactly, with a Mermaid flowchart for paths of up to 12 steps.

Example:

User: "How are Ruth and Jesus related?"

Claude calls: find_connection(person1="Ruth", person2="Jesus")

Returns: Ruth → child → Obed → child → Jesse → child → David → ... → Jesus

15. people_in_passage

Find all people, places, and events mentioned in a Bible passage.

What it returns: All entities mentioned in a chapter or verse according to Theographic Bible Metadata.

Cultural Context

18. get_ane_context

Retrieve Ancient Near East cultural and historical context that illuminates biblical passages. Covers 12 dimensions of ANE life across 9 time periods.

When Claude uses it: User asks about cultural background, customs, practices, or worldview behind a passage; when understanding the original context would prevent modern misreadings.

What it returns: Structured entries with title, summary, detailed explanation, ANE parallels from extra-biblical sources, interpretive significance, key references, and scholarly sources.

Dimensions: Cosmology & Worldview, Religious Practices, Social Structure, Legal & Covenant, Political & Imperial, Economic Life, Literary Conventions, Warfare & Military, Daily Life & Material Culture, Death & Afterlife, Gender & Family, Education & Literacy.

Periods: Patriarchal, Exodus/Conquest, Judges/Early Monarchy, United Monarchy, Divided Monarchy, Assyrian/Babylonian, Persian, Hellenistic, Roman.

Example:

User: "What's the cultural background behind Genesis 15 - the covenant ceremony?"

Claude calls: get_ane_context(reference="Genesis 15:1", dimension="legal_covenant")

Returns:
- Suzerainty Treaty Forms: ANE covenant structure (preamble, stipulations,
  witnesses, curses/blessings) parallels Deuteronomy and Genesis 15
- Self-Maledictory Oath: Walking between animal halves = "may I become
  like these animals if I break this covenant"
- ANE Parallels: Hittite suzerainty treaties, Mari covenant rituals
- Interpretive Significance: God alone passes through the pieces,
  taking the covenant curse entirely upon himself

Hybrid Search

16. graph_enriched_search

Combine passage lookup with graph context for the most comprehensive single-query view of a verse.

What it returns: Verse text (English + original), all people/places/events mentioned, family relationships for each person mentioned.

17. find_similar_passages

Discover semantically similar passages across the Bible using vector embeddings.

When Claude uses it: User wants to find thematic connections not captured by explicit cross-references, or asks "What passages are similar to X?"

What it returns: Similar passages ranked by semantic similarity (%), genre warnings, hermeneutical cautions.

Important: Semantic similarity indicates shared vocabulary and concepts, but does NOT establish theological connection. The tool includes warnings to verify genre compatibility, authorial intent, and historical context.

Example:

User: "Find passages similar to Daniel 7:7"

Claude calls: find_similar_passages(reference="Daniel 7:7", limit=5)

Returns:
- Daniel 7:15-28 (70.0% similar) - interpretation of the vision
- Revelation 9:17-21 (61.1% similar) - locusts with lion's teeth
- Daniel 2:29-45 (60.8% similar) - Nebuchadnezzar's statue

Theological Scholarship & Structure

get_theology_context

Retrieve attributed theological scholarship by biblical reference, theme, or author.

get_torah_weave

Find structurally paired Torah passages under Moshe Kline's Woven Torah hypothesis.

get_textual_variant

Inspect catalogued MT/LXX/DSS divergences without collapsing sources and conclusions. Each reading lists only the manuscripts that witness it. New Testament quotations appear under NT citation evidence, and material that bears on a text form without witnessing the verse (such as a Qumran scroll attesting a short edition elsewhere in Jeremiah) appears under Indirect evidence. Corpora with no surviving copy are listed in a case-level Not preserved in note. Every witness keeps the dataset's raw support label.

Critical judgments contain only conclusions that carry a source citation. Uncited dataset summaries appear separately as an Unsourced dataset summary (not verified against published text-critical scholarship). Theological interpretations, named-author claims, and NT quotation alignment also stay in their own sections, and sections with nothing catalogued are omitted. The dataset's Heiser Literal Translation (HLT) renderings and rationales stay in the database but are not printed: the HLT is the owner's own unpublished translation. Verses where the witnesses agree with the MT but the reading or rendering is contested (for example the divine-council language of Psalm 82:6) appear as Translation and interpretation notes, never as variants. A verse can have both: at Job 38:7, Daniel 4:17 and 7:13, and Psalm 29:1 a variant case records where the Greek differs from the MT, and the notes follow it; Mark 14:62, Matthew 24:30 and 26:64, and Revelation 1:7 reach the Daniel 7:13 case through NT quotation alignment. lookup_verse mentions when a verse has a catalogued variant case or interpretation notes. On a database that still has the legacy variant schema, the tool asks you to run python scripts/upgrade_database.py.

Further Research

list_research_sources

List curated, freely accessible research repositories to consult before a general web search: primary-text libraries (Sefaria, CCEL, Perseus, the Internet Archive), open-access journals (Tyndale Bulletin, Themelios, the Journal of Hebrew Scriptures, HTS), Dead Sea Scrolls and Septuagint resources, and Heiser's own sites. Each entry states its coverage, perspective (Jewish, Catholic, Reformed, evangelical, academic/critical, multi-tradition, single-author) and access or licence terms. Filter by topic (coverage tag or name, matched loosely) and perspective; pass query to get a ready-to-open search link for every source that supports one (STEP Bible's link opens the passage when the query is a Bible reference). With no filter the list starts with a mix: Tyndale Bulletin (evangelical), STEP Bible (reference), Sefaria (Jewish) and Bible Odyssey (academic). The list is a set of places to look, not endorsements.

The catalogue lives in src/study_bible_mcp/research_sources.json. It is a directory of repositories, not a per-passage or per-scholar index.

Checking links. python scripts/check_links.py fetches, over the network, every catalogue URL, every search template filled with a sample query, and (when db/study_bible.db or --db is present) every distinct theology_content.url in the database. It makes about 100 requests to third-party sites, so run it occasionally rather than in CI. Checks run with bounded concurrency and a 20-second timeout; a link that times out is retried once with a 60-second limit, because some hosts are slow rather than dead. Any link answering with status 400 or above, or failing at the network level, is reported as BROKEN with where it came from; one that still times out after the retry is reported as TIMEOUT. Either makes the run exit non-zero. Hosts known to be slow but working are listed in LINK_CHECK_SLOW_HOSTS in the script (currently drmsh.com, which has taken 31-57 seconds to answer): their retry gets a 180-second limit, and a timeout that still persists is reported as SLOW, a warning that does not fail the run. A real error status from those hosts still does. Some sites refuse HEAD requests, so the checker falls back to GET. Some answer every automated client with a bot challenge; only for those, an entry may set "link_check": "skip" with a "link_check_reason", and the checker lists it as skipped rather than checking it. Verify skipped entries by hand when updating last_verified.

How Users Interact

Natural Language Queries

Users don't need to know the tool names. They just ask questions:

Example Conversation

User: "I'm studying Romans 5:8. What does Paul mean when he says God 'demonstrates' his love?"

Claude's internal process:

1. Look up the verse: lookup_verse(reference="Romans 5:8")
2. Study the key word: word_study(word="demonstrates") → finds G4921 συνίστημι
3. Get study notes: get_study_notes(reference="Romans 5:8")
4. Check cross-references for context
5. Apply genre guidance (Epistles)

Claude's response:

In Romans 5:8, Paul uses the Greek word συνίστημι (synistēmi, G4921), which means "to stand together, to commend, to demonstrate." This is a present tense verb, indicating ongoing action.

The genre context is important here: Romans is an epistle, and Paul is building an argument. In the immediate context (Romans 5:1-11), he's explaining how we have peace with God through justification by faith. Verse 8 is his proof: God didn't wait for us to become worthy, but demonstrated his love "while we were still sinners."

The word "demonstrates" carries the sense of proving beyond doubt - not just telling us about his love, but showing it through action (Christ's death).

Related passages on God's love demonstrated through Christ: John 3:16, 1 John 4:9-10.

Example Questions

Try asking Claude:

What does John 3:16 say in Greek? Break down the key words.

Study the word 'agape' - how is it different from other Greek words for love?

What's the context of Romans 8:28? Who is Paul writing to and why?

Show me cross-references for Ephesians 2:8-9 on salvation by grace.

Who was Melchizedek and why is he significant?

Parse the verb morphology in Philippians 2:12 - "work out your salvation"

What does the Hebrew word 'hesed' mean? Where is it used?

Help me understand Revelation 13 - what are these beasts?

What do the study notes say about John 1:1?

Look up 'baptism' in the Bible dictionary.

Show me David's family tree going back 5 generations.

How are Abraham and Jesus connected genealogically?

What events happened in the life of Moses?

Find passages similar to John 3:16 - what other verses talk about God's love like this?

What Old Testament passages connect to Daniel's vision of the Son of Man?

What's the Ancient Near East background for the creation account in Genesis 1?

Explain the cultural context of covenant-making in Genesis 15.

Hermeneutical Framework

This isn't just a reference tool. It's designed to guide users toward sound biblical interpretation.

The Problem It Solves

Common interpretation errors include:

• Proof-texting: Using isolated verses to support predetermined conclusions
• Allegorizing: Finding hidden meanings not intended by the author
• Ignoring genre: Reading poetry as prose or prophecy as history
• Ignoring context: Ripping verses from their historical and literary setting
• Moralizing: Reducing narratives to "be good like David"

The Fee & Stuart Method

The interpretive framework is based on "How to Read the Bible for All Its Worth" by Gordon Fee and Douglas Stuart - the standard textbook for biblical hermeneutics used in seminaries worldwide.

About the authors:

• Gordon Fee (1934-2022): New Testament scholar, professor at Regent College, author of major commentaries on 1 Corinthians, Philippians, and Pauline theology. One of the translators of the NIV and TNIV.
• Douglas Stuart: Old Testament scholar, professor at Gordon-Conwell Theological Seminary, author of commentaries on Hosea-Jonah, Exodus, and Old Testament hermeneutics.

Their book (first published 1981, now in its 4th edition) has sold over a million copies and is used as the introductory hermeneutics text at most evangelical seminaries. It teaches readers to interpret Scripture according to its literary genre, historical context, and authorial intent - avoiding both wooden literalism and uncontrolled allegorizing.

The server embeds genre-specific interpretation principles from this methodology:

Core Principles

1. Exegesis Before Application

   • First discover what the text meant to its original audience
   • Then determine what it means for us today
   • Never jump directly to application

2. Context is King

   • Historical context: time, culture, author, audience, occasion
   • Literary context: genre, surrounding passages, book structure
   • A text without context is a pretext for a proof text

3. Genre Determines Method Different biblical literature requires different approaches

Genre-Specific Guidelines

Epistles (Letters)

Key principle: Think contextually - what problem was being addressed?

• Read the entire letter to understand the argument flow
• Ask: What problem prompted this letter?
• Pay attention to logical connectors (therefore, because, so that)
• Distinguish between central argument and supporting points
• Not every statement is universally prescriptive

Common errors:

• Taking verses out of their argumentative context
• Ignoring the specific historical situation being addressed
• Treating every instruction as a timeless command

Old Testament Narratives

Key principle: Three levels (meta, national, individual); descriptive not prescriptive

• Narratives describe what happened, not necessarily what should happen
• Three levels: God's universal plan, Israel's story, individual accounts
• Characters are not always examples to follow - observe their flaws
• The narrator rarely makes direct theological statements
• God is the ultimate hero of every narrative

Common errors:

• Treating every character's action as a model to follow
• Allegorizing details that are simply historical facts
• Moralizing: "Be brave like David" without seeing God's work

Gospels

Key principle: Two-level documents (Jesus' original context + evangelist's purpose)

• Consider both Jesus' original meaning and the Gospel writer's purpose
• Each Gospel has a specific audience and theological emphasis
• Account for the already/not-yet tension of the Kingdom
• Compare parallel accounts to see each writer's emphasis

Common errors:

• Harmonizing too quickly without hearing each Gospel's voice
• Ignoring Jesus' audience (Pharisees, disciples, crowds)
• Applying teachings directly without considering the cross

Parables

Key principle: Find the main point; don't allegorize every detail

• Identify the one or two main points - parables are not allegories
• Consider the audience: Who is Jesus speaking to and why?
• Look for cultural background that illuminates meaning
• The unexpected twist often carries the main message

Common errors:

• Allegorizing every detail (the donkey represents X...)
• Missing the shock value for the original audience
• Creating theological systems from parabolic details

Prophets

Key principle: Covenant enforcement; check if "future" is now past

• Prophets were covenant enforcement officers - calling Israel back
• Most "predictions" were about events now in our past
• The prophetic lawsuit pattern: accusation, judgment, hope
• Distinguish near fulfillment from ultimate/eschatological fulfillment
• Poetry and metaphor are the normal mode of prophetic speech

Common errors:

• Reading all prophecy as about the end times
• Literalizing poetic and metaphorical language
• Creating detailed end-times timelines from prophetic poetry

Psalms

Key principle: Poetry/prayer; understand the types (lament, thanksgiving, praise)

• Psalms are inspired responses to God - poetry, not doctrine
• Identify the type: Lament, Thanksgiving, Praise, Royal, Wisdom
• Poetry uses metaphor, hyperbole, and parallelism
• The emotions are real and God-given - even the difficult ones
• Many psalms find their ultimate fulfillment in Christ

Common errors:

• Treating poetic expressions as doctrinal statements
• Ignoring the emotional dimension
• Literalizing metaphorical language

Wisdom Literature

Key principle: General truths, not guarantees; compare with full canon

• Proverbs are general truths, not unconditional promises
• Wisdom literature wrestles with life's complexity
• Job and Ecclesiastes challenge simplistic interpretations
• The fear of the Lord is the foundation of all wisdom
• Compare individual proverbs with the full biblical witness

Common errors:

• Treating Proverbs as unconditional promises
• Ignoring the dialogue format in Job (not all speeches are true)
• Missing the "under the sun" perspective in Ecclesiastes

Apocalyptic (Revelation)

Key principle: Apocalyptic imagery; already/not yet eschatology

• Apocalyptic is a literary genre with its own conventions
• Symbols are stock imagery (beasts, numbers) with recognized meanings
• The primary message is: God wins, evil is judged, Christ reigns
• Read in light of OT prophets and the already/not yet tension
• Multiple valid interpretive approaches exist

Common errors:

• Literalizing symbolic imagery (666 as a barcode)
• Creating detailed chronological timelines
• Missing the pastoral purpose: encouragement for suffering believers

The Reasoning Pattern

The server's hermeneutics module guides Claude through this sequence:

1. IDENTIFY: What biblical text(s) are relevant?
2. GENRE: What type of literature is this?
3. CONTEXT: Historical, literary, canonical
4. CONTENT: What does the text say? (Greek/Hebrew when helpful)
5. CROSS-REFERENCES: Scripture interprets Scripture
6. APPLICATION: Timeless principle behind temporal commands
7. HUMILITY: Acknowledge uncertainty where it exists

Self-Hosting

Want to run your own instance? See docs/SELF_HOST.md.

Options:

• Local Python development
• Docker deployment
• Your own Fly.io instance

Project Structure

studybible-mcp/
├── src/study_bible_mcp/
│   ├── server.py              # MCP server (stdio, SSE, and Streamable HTTP)
│   ├── database.py            # SQLite queries (async)
│   ├── tools.py               # MCP tool definitions and output formatters
│   ├── research_sources.json  # Curated open research repositories (list_research_sources)
│   ├── hermeneutics.py        # Genre detection & interpretation
│   ├── corpora/               # Ancient-source corpus registry
│   │   ├── __init__.py        # CORPORA registry (one line per corpus)
│   │   ├── base.py            # Corpus, Work, Collection, BiblicalLink types
│   │   ├── locators.py        # Citation locator grammars and sort keys
│   │   ├── citations.py       # Citation-to-work resolution
│   │   ├── download.py        # Checksum-verified download helpers
│   │   ├── josephus.py        # Josephus TEI parser, works, and links
│   │   ├── targum.py          # Targum Onkelos Sefaria files, verse keys
│   │   └── rabbinic.py        # Mishnah (Sefaria)
│   └── parsers/
│       ├── lexicon.py         # TFLSJ, BDB, Abbott-Smith, TBESG, TBESH parsers
│       ├── aquifer.py         # BibleAquifer JSON parser
│       ├── acai.py            # ACAI entity annotation parser
│       ├── ane_context.py     # ANE cultural context parser
│       ├── biblical_versions.py # MT/LXX text, morphology, and versification
│       ├── bsb.py             # Berean Standard Bible (bsb.txt) parser and importer
│       ├── ancient_sources.py # Generic corpus importer (passages, FTS, links)
│       ├── tagged_text.py     # TAGNT/TAHOT morphology parsers
│       └── proper_names.py    # TIPNR name parser
├── scripts/
│   ├── download_stepbible.py  # Download base source data (--aquifer flag)
│   ├── download_biblical_versions.py # Download pinned LXX/Brenton sources
│   ├── download_bsb.py        # Download the pinned Berean Standard Bible text
│   ├── download_ancient_sources.py # Download every registered ancient corpus
│   ├── build_database.py      # Build SQLite database
│   ├── generate_embeddings.py # Generate vector embeddings
│   ├── import_theographic.py  # Import Theographic graph data
│   ├── upgrade_database.py    # In-place schema/source upgrade of an existing DB
│   ├── smoke_server.py        # Full-database smoke suite
│   └── check_links.py         # Checks research-source and theology URLs (network)
├── tests/                     # Deterministic unit and tool-selection tests
├── prompts/
│   └── system_prompt.md       # Full hermeneutical framework for agents
├── docs/
│   ├── SETUP.md               # Quick setup guide
│   └── SELF_HOST.md           # Self-hosting instructions
├── ARCHITECTURE.md            # Mermaid flowchart of the tool-selection paths
├── Dockerfile                 # Runtime image; database mounted separately
├── fly.toml                   # Fly.io deployment config
└── pyproject.toml

Data Sources

STEPBible - Tyndale House, Cambridge

Biblical text and brief lexicon data come from the STEPBible project (GitHub), developed by scholars at Tyndale House, Cambridge - one of the world's leading centres for biblical research. Licensed CC BY 4.0.

Berean Standard Bible

The English verse text shown by lookup_verse, search_by_strongs, word_study, find_similar_passages and graph_enriched_search is the Berean Standard Bible, which its publisher has dedicated to the public domain. scripts/download_bsb.py fetches bsb.txt from bereanbible.com, which is unversioned, so the file is pinned by SHA-256 (BSB_SHA256 in src/study_bible_mcp/parsers/bsb.py, the edition downloaded on 2026-10-14); the download and import fail on any other file. The import also requires the file's own public-domain statement, and moves the Psalm headings the file prints at the start of verse 1 into verse 0 rows, so a Psalm's superscription shows its BSB heading and verse 1 is the verse alone. Verses are matched by English numbering, so split-numbered STEPBible rows (Gen.32.1(32.2)) get their BSB verse; where the BSB has no such verse the tools show the STEPBible gloss marked [gloss], and without the table they show the gloss as before.

Septuagint and MT versification

The LXX comparison corpus uses the complete 59-book OpenScriptorium lxx-morph Rahlfs morphology dataset, pinned to commit c91f6b1e8fb3ba37df701e6ae31f675ace71a2b2. Its explicit verse_pairs.jsonl mapping resolves Psalm numbering, Jeremiah's chapter order, 2 Esdras, and smaller MT/LXX shifts. That mapping is derived in part from TVTMS (Tyndale Versification Traditions Mapping System), CC BY 4.0, created by STEPBible.org / Tyndale House Cambridge. Downloads are checked against the pinned Git blob hashes. Morphology is CC BY 4.0; the underlying Rahlfs (1935) Greek text is public domain. Only the LXX rows are stored (about 35 MB including indexes, with word morphology zlib-compressed); MT rows are read from the existing TAHOT verses table at query time.

English LXX translations come from the public-domain Brenton Septuagint at eBible.org, and Brenton's own public-domain Greek text from eBible's grcbrent (data/lxx/brenton-greek). eBible regenerates its unversioned zips, so the download script verifies the SHA-256 of every extracted USFM book rather than of the archive, and never deletes or overwrites local files that fail verification. Brenton's Greek is stored only for the verses where it differs from Rahlfs (about 5 MB). build_database.py stops with an error if data/lxx is present but incomplete (any of the 59 morphology books, the mapping, a required Brenton English or Greek book, or a manifest-listed file is missing or truncated).

Ancient sources

The ancient sources are texts written by about AD 200 that witness the Second Temple world: Josephus, Targum Onkelos, four Qumran texts (CD, 1QS, 1QSa, 1QpHab), Philo, 1 Enoch, 4 Ezra and the Mishnah. Targum Onkelos is the one stated exception: its final form is c. 3rd–5th c. AD, and it is kept for its earlier tradition as the standard Torah targum; its tag gives the final-form date and its caveat states the exception. Sources later than c. AD 200 (Targum Pseudo-Jonathan with Etheridge's English, the Tosefta, the Babylonian Talmud and Leviticus Rabbah) were removed: they are too far from the New Testament to witness its world without an earlier source.

Each corpus is a module in src/study_bible_mcp/corpora/ registered in CORPORA; its pinned files live in data/ancient_sources/<corpus>/ and source_manifest.json records every verified file. build_database.py imports every corpus whose directory exists and fails if a present directory is incomplete; upgrade_database.py fails unless every registered corpus is downloaded and complete. All corpora share ancient_source_passages, whose sort_start/sort_end keys (in each work's citation grammar) back lookup_ancient_source, and whose original_license, translation_license, and combined license record provenance per passage. Each import replaces only its own corpus inside a savepoint after validating every passage and link, and merges the FTS index so repeated upgrades do not grow it. Indexed originals are folded for search: Greek accents and breathings and Hebrew/Aramaic vowel points and cantillation are removed and case is folded, in the index and the query. Editorial signs inside words ([ ] ⟨ ⟩ < > { } ( ) ^ ¦) and invisible format characters are deleted rather than treated as word breaks, in the index and the query; a corpus can also declare search_index_removals, patterns for printed text that is not part of the reading (Qumran's scribal erasures and vacat), which are deleted from the index, and search_index_gaps, printed text that is not evidence (Qumran's [ ] reconstruction), which becomes a word break, so it is never matched and the surviving letters either side are not joined into a false word. Every Hebrew-script word is also indexed, in a separate column searched only as a fallback, without one or two leading prefix letters, keeping at least three letters. The displayed text keeps every sign.

Every passage is labelled so that the agent never mistakes it for Scripture:

• Tag. Each corpus (or work) declares a SourceTag in the registry: canon status, kind, composition date, date tier (A: written before or alongside the NT; B: after the NT, to c. AD 200, with Targum Onkelos's later final form the stated exception), edition, and, where the copy is much later than the work, the manuscript date. 1 Enoch is dated book by book. The registry refuses a corpus with an untagged work.
• Caveat. Each corpus declares a 1–3 sentence caveat, printed once per response that shows it (e.g. Philo: his Logos is not the incarnate Son).
• Editorial notes. src/study_bible_mcp/corpora/editorial_notes.json (committed) holds verse-level notes — emendation, interpolation, reconstruction, contrast, polemic, caution, gap — with contrasting and aligned biblical references, and extra typed links. Each corpus import loads them into ancient_source_notes (one row per passage a citation covers; a gap note sits on the passages either side of its range) and fails if a citation covers no passage or a biblical reference does not parse. A file link to the same passage and range as a module link replaces it. Notes print as ⚠ <note> (Contrast: Mark 14:62), once per response. The file holds 46 notes and 40 links, e.g. Charles's emendation at 1 Enoch 71:14, the Christian "Jesus" of 4 Ezra 7:28–29, the restored lemma at 1QpHab 7:17, Philo's Logos at Conf. 146, and Yom Kippur atonement at m. Yoma 8:8–9.
• Typed links. ancient_source_biblical_links.relation is aligned, echo, shared_tradition, contrast, or translation, printed as Aligned: Jude 14–15 — …, Contrast: … and so on. A pseudepigraphon is paired with the Scripture that takes it up (1 Enoch with Gen 6:1–4, Jude 6 and 14–15, 2 Pet 2:4, 1 Pet 3:19–20). Module links take echo when their note says the text cites, quotes or expounds the Bible and shared_tradition otherwise, unless they set a relation; create_schema adds the column to older databases with the same rule, and rebuilds an older search index.

Josephus

The josephus corpus uses all four Josephus works from PerseusDL canonical-greekLit, pinned to commit 01b725d835e6e733062ffd79e0efdbae1ba06e5c. The English TEI has one division per Whiston section, labelled with the Niese section where it starts; the Greek TEI has one division per Niese section. The importer joins every Greek section from each English start up to the next (the last section of a book runs to the book's last Greek section), so no Niese Greek is dropped; the later Greek book summaries (arg) are excluded. One mislabelled English division (Whiston 17.13.1, labelled 13) is corrected to Niese 17.339 from the Greek milestone and text. Editor notes and Whiston's chapter headings are kept out of the primary-source text and the search index; headings are stored as editorial metadata. The FTS5 index uses Porter stemming for English and an accent-folded copy of the Greek. Each TEI file declares CC BY-SA 4.0; downloads are checked against their pinned Git blob hashes, and the build fails if any of the eight TEI files is missing.

Biblical links are a small, separate editorial index of direct narrative overlap. Each targets a Niese section, attaches to the passage whose span contains it, and was checked against the passage text; the import fails if a link does not resolve. They do not assert literary dependence or historical agreement, and the Testimonium Flavianum link is marked as of disputed authenticity.

Targums

The targum corpus holds Targum Onkelos on all five books of the Torah, one work per book (onkelos_genesis … onkelos_deuteronomy, cited as Onkelos Genesis 1:1 or Tg. Onq. Deut 32:8). The 10 version files come from Sefaria's archived export, Sefaria/Sefaria-Export-Archive at commit 3f1013631fdfe452e953a93a2c5f921319e394ed (the last commit holding the JSON texts), each checked against its pinned Git blob hash and against the license and versionTitle it declares; merged.json files are never used.

• Aramaic: the Torat Emet text of Onkelos (Sefaria versions Onkelos {Book}), declared Public Domain by Sefaria. One passage per Sefaria verse (5,846). A ??? digitization mark before Genesis 49:12 is removed.
• English: the Metsudah Chumash (Metsudah Publications, 2009, Sefaria version Metsudah Chumash, Metsudah Publications, 2009 [with Onkelos translation], CC BY-NC 4.0). It translates the Hebrew text, so it is not a direct translation of the Aramaic beside it. Metsudah sets Onkelos's own renderings in bold (in Genesis 12–17, italic) brackets; these print as ⟨…⟩ (3,599 of them), while plain [ ] are Metsudah's own supplements, e.g. Genesis 49:10 "until Shiloh Comes ⟨the Moshiach will come, for his is the kingship⟩". The tag says "Metsudah: English of the Hebrew Torah; ⟨marked⟩ words are Onkelos's renderings", and the licence line reads Aramaic text: Public domain; English translation: CC BY-NC 4.0. A verse without Metsudah English is an Aramaic-only passage.

Versification. Sefaria numbers Onkelos by the Hebrew verses; passages, citations and links use the server's English numbering, with the Hebrew number in the heading where it differs (Targum Onkelos Exodus 8:5 (MT 8:1)). A static table in targum.py covers the Torah's 18 differing runs, as in TAHOT's verses table (Exo.8.5(8.1)): Gen 31:55–32:32, Exod 8:1–32 and 22:1–31, Lev 6:1–30, Num 16:36–17:13 and 29:40–30:16, Deut 12:32–13:18, 22:30–23:25 and 29:1–29. Sefaria's Decalogue follows the upper accents, so Exodus 20:13–16 and Deuteronomy 5:17–20 are one passage each and later verses shift by three; Numbers 26:1 also holds MT 25:19. The import fails if the English keys stop rising.

Each passage has one editorial "Translation of" link to the English verse(s) it covers, printed with the Hebrew number where it differs (Translation of: Exod 8:5 (MT 8:1)), so biblical_reference finds Onkelos on a verse. The period tag gives the redaction era (editorial): amoraic.

Qumran

The qumran corpus has one passage per manuscript column, its Hebrew lines prefixed with their line numbers and cited column:line. No English translation is included, so every passage is original-only and the tools say so.

• Damascus Document, Community Rule, Rule of the Congregation, Pesher Habakkuk (CD, 1QS, 1QSa, 1QpHab): the Text-Fabric transcription by Martin G. Abegg Jr., James E. Bowley and Edward M. Cook in ETCBC/dss, release 2.0.1, pinned to commit 47ecc1738fad6e67df7d40a3e84d73e4a125e265 and licensed CC BY-NC 4.0 (checked in every feature file's header at download and import). CD keeps the conventional page numbers: Genizah MS A columns 1–16 and MS B columns 19–20, shown as CD-A/CD-B alongside, and its tag names the medieval copy served (copy: Cairo Genizah MS A, 10th c. AD). Lines are rendered from the sign-level features with the transcription's brackets: [ ] modern reconstruction, { } and {{ }} removal by a modern editor or the scribe, < > and << >> correction by a modern editor or the scribe, ^ ^ supralinear, ( ) alternative reading, and vacat (spaced off from the text) for an unwritten space. A circlet above a letter (U+05AF) marks any of its four uncertainty degrees, ◦ an unidentified sign, and … a missing sign; the morpheme breaks Abegg inserts before suffixes are dropped, and the transcription's ׃ marks are kept. Search ignores the circlets and indexes the corrected, surviving reading: modern reconstruction in [ ] is never matched (the letters either side of יג[די]לו are not joined into יגלו), and letters in { } and {{ }} and the word vacat are left out of the index (י{{א}}<<ע>>בורו is found as יעבורו); the displayed text keeps every sign. Other ETCBC scrolls are not imported (1QSb, for example, mixes fragment labels and line gaps).

The 32 biblical links connect a 1QpHab column to each Habakkuk 1–2 verse (range) whose lemma it quotes. Each was checked against the Masoretic verse, and the import fails unless the anchor words survive outside the editor's [ ] on their line. Three say how much of the lemma is restored; 7:14 links Hab 2:4a only, "(end of 2:4a restored; 2:4b wholly reconstructed)".

R. H. Charles's public-domain English of the Damascus Document ("Fragments of a Zadokite Work", APOT II, 1913) is not used. It is available only as OCR, in Charles's own chapter:verse numbering with no Schechter page and line references, and his chapter IX prints MSS A and B in parallel columns that the OCR interleaves line by line, so it cannot be aligned reliably with CD lines.

Philo

The philo corpus uses the Greek (grc1, Cohn–Wendland–Reiter, Philonis Alexandrini opera quae supersunt, Berlin: Reimer 1896–1915) and English (eng1, C. D. Yonge, The Works of Philo Judaeus, London: Bohn 1854–55) TEI of all 31 treatises in OpenGreekAndLatin/First1KGreek (data/tlg0018/tlg001–tlg031, De Opificio Mundi to Legatio ad Gaium), pinned to commit 03776b39f4047c5cff06f5296fae4b2bae4b08fb. The module pins each file's Git blob hash and fetches it from that commit without GitHub API calls. The First1KGreek editors divided the English at the same Cohn–Wendland sections as the Greek, so each passage is exactly one section: its Greek and the English section of the same number, never a neighbouring one. The 58 Greek sections of Spec. 1.177–193 and 2.124–213 that Yonge's shorter text lacks are imported as Greek-only passages, marked "Yonge has no English for this section"; an English section without a Greek counterpart fails the import. Apparatus, marginal notes, page and line markers, and Yonge's footnotes and source references are kept out of the text; Yonge's treatise headings (e.g. "A TREATISE ON CIRCUMCISION." within Spec.) are editorial headings. Each file declares CC BY-SA 4.0, checked at download and parse. Works surviving only in Armenian or Latin (QG, QE, Prov., Anim.), the Hypothetica, and fragments are not in First1KGreek and are not included. Twenty-five curated links name the Genesis or Exodus verse a section quotes and expounds (e.g. Leg. 1.31 → Gen 2:7) and Legat. 299 → Luke 3:1 (Pilate). Search for Philo's Greek uses Greek script (λόγος, λογος); a transliteration such as logos does not match, and Yonge renders λόγος as "word" or "reason".

Pseudepigrapha

The pseudepigrapha corpus holds two non-canonical works. 1 Enoch is R. H. Charles's translation (2nd ed. 1912, SPCK reprint 1917) from English Wikisource, one page per chapter (1–108), each pinned to a revision id and its SHA-1 and fetched in three MediaWiki API requests. The translation is in the public domain in the United States (Charles died in 1931; Wikisource {{PD/US|1931}}). Each passage is one Charles verse in his order, including his sub-verses and transpositions (1 Enoch 5:6a–b, 5:7c, 5:6d–j; 89:48b; 91:14d), all sorted under their verse number; Charles's section headings (LXXXIX. 72-77. ...) are editorial headings. Where Charles prints a parallel column with a header (his rendering of the Greek or Latin witnesses in chapters 22, 27, 32, 89, and 106) only the main Ethiopic column is imported; the unheaded doublet columns of 90:13–18 are both text. No Ge'ez is included: the Beta maṣāḥǝft file (LIT1340EnochE) offers Dillmann's text only at chapter level under a custom non-commercial notice that must accompany every copy, and an EOTC printed-edition text whose own rights are not stated.

4 Ezra is 2 Esdras 3–14 in the KJV with Apocrypha (eBible.org eng-kjv, public domain; the extracted 58-2ESeng-kjv.usfm is pinned by SHA-256 because eBible regenerates its zip) paired with the Clementine Vulgate's Liber Quartus Esdrae (Rome 1598) from Latin Wikisource revision 148869 (transcribed from public-domain page images; Wikisource transcription CC BY-SA 4.0, credited by URL in every Sources block). Passages are one KJV verse with the Latin verse of the same number, except 27 spans where the two divide verses differently (usually a speech introduction such as "Et dixit ad me," / "And he said unto me" placed on the other side of a verse boundary, and an offset in 14:14–20); each span pairs all of its English with all of its Latin, and the import fails if any verse of either text is left unpaired. 4 Ezra 3–14 keeps the 2 Esdras chapter numbers; KJV 7:36–70 is cited as 7:106–140 (both texts lack the "missing fragment" 7:36–105, which is not available from an openly licensed source). 2 Esdras 1–2 and 15–16 are later Christian additions (5 and 6 Ezra) and are not included. A citation outside the served text (4 Ezra 7:50, 2 Esdras 15:1) says both.

Period tags give conventional composition dates (1 Enoch: Watchers, Astronomical Book, Dream Visions, and Epistle hellenistic, Parables roman; 4 Ezra roman, after 70 CE). Twenty-two curated links mark allegorical retellings and shared imagery, e.g. the Animal Apocalypse (1 Enoch 89) with the Genesis–Kings narrative and Malachi 1:7, and the eagle vision (4 Ezra 11–12) with Daniel 7, which 4 Ezra 12:11 names; 1 Enoch 1:9 → Jude 14–15 and 6:1 → Gen 6:1–2 are aligned. The editorial-notes file adds the aligned links to Jude 6, 2 Pet 2:4 and 1 Pet 3:19–20.

Rabbinic texts

The rabbinic corpus holds the Mishnah from Sefaria's JSON export, pinned to commit 3f1013631fdfe452e953a93a2c5f921319e394ed of Sefaria/Sefaria-Export-Archive (the last full export committed before Sefaria moved it to a GCS bucket that is overwritten monthly and not versioned) and verified by Git blob hash. Every Sefaria version declares its own licence; downloads and imports reject any version that is not Public Domain or CC-BY. CC-BY is recorded as CC BY 3.0, the licence Sefaria's reader links it to.

• Mishnah, all 63 tractates, one passage per mishnah (Mishnah Shekalim 1:1): Hebrew Mishnah, ed. Romm, Vilna 1913 (public domain) with Joshua Kulp's Mishnah Yomit English (CC BY). Bikkurim chapter 4, absent from the Vilna version, is English only.

Footnotes, footnote markers and HTML are removed from the English. Curated links: Mishnah Shekalim 1:1 and 1:3 to Matthew 17:24–27 (the half-shekel); the editorial-notes file adds Gospel echoes (e.g. Shekalim 1:3 → Mark 11:15) and contrast notes. The corpus caveat warns that Kulp's bracketed Bible references sometimes use Hebrew numbering or contain errors.

BDB Hebrew Lexicon

The full unabridged Brown-Driver-Briggs Hebrew lexicon (8,090 entries) comes from eliranwong/unabridged-BDB-Hebrew-lexicon. The BDB is the standard scholarly Hebrew-English lexicon, originally published in 1906 and still widely used in academic study. Public domain text.

Abbott-Smith Greek Lexicon

G. Abbott-Smith's A Manual Greek Lexicon of the New Testament (1922) provides NT-focused definitions that complement the broader classical coverage of LSJ. The TEI XML edition (5,896 entries) comes from translatable-exegetical-tools/Abbott-Smith. Public domain text.

Abbott-Smith enriches 5,426 existing Greek entries with:

BibleAquifer

Scholarly commentary and reference content from BibleAquifer (CC BY-SA 4.0):

ACAI Entity Annotations

Rich entity annotations from BibleAquifer/ACAI (CC BY-SA 4.0):

Theographic Bible Metadata

Graph data for genealogy, events, and places from Theographic, with curated corrections and twenty added events (the binding of Isaac, the burning bush, the fall of Jerusalem ...) in src/study_bible_mcp/parsers/graph_curated.json. Used by the explore_genealogy, explore_person_events, explore_place, find_connection, people_in_passage, and graph_enriched_search tools.

Torah Weave

get_torah_weave reads version 1.1 of Moshe Kline's Torah Literary Units dataset (chaver.com, CC BY 4.0; Kline, Before Chapter and Verse: Reading the Woven Torah, 2022): 86 units of Genesis–Deuteronomy with 600 cells, 1,019 counting subdivisions. The file ships as data/torah_weave/torah-units.json. Its URL is unversioned, so it is pinned by SHA-256 (TORAH_WEAVE_SHA256 in src/study_bible_mcp/parsers/torah_weave.py, the file retrieved on 2026-10-14), and the import fails on any other file. Kline numbers verses like the Hebrew Bible (Leviticus 5:20-26 is English 6:1-7); the tool takes and shows English numbers. Every Torah verse is in a cell except Deuteronomy 5:30-33, which sit in unit 76 without one, and Exodus 20:23-26, which no unit includes.

Ancient Near East Context

87 structured entries covering 12 dimensions of ANE cultural and historical context across 9 time periods, with 314 book-chapter mappings. Data is stored in data/ane_context/ as JSON files (one per dimension). Sources include Walton, Hallo, Matthews & Benjamin, Pritchard (ANET), and other standard ANE reference works.

Theological Scholarship

193 verse-anchored entries covering Heiser, Bradley, Burnett, Owen, Stott, and Lennox, served by get_theology_context.

These entries are original summary prose written for this database, describing each author's argument on a passage or theme with attribution. They are not reproduced text from the authors' works. The material was derived from publicly available research and teaching — papers, articles, lectures, and podcasts — rather than from in-copyright books; where a specific public source is known, the entry carries its URL and the tool output shows it. Every result from get_theology_context states this provenance inline.

John Owen (d. 1683) is in the public domain, so those entries are cited directly by work.

Source links

An entry's url is shown as Source URL in get_theology_context output. Links are added only where the source itself is legitimately free online:

In the content JSON, a file-level url applies to every entry; an entry-level url (for example a chapter page) overrides it.

Applying data changes to an existing database

The importers are re-runnable. A row that already exists (same source_work, chapter_or_episode, title) is not duplicated and keeps its content; only its url is refreshed. To apply new or corrected links to an existing DB:

DB=db/study_bible.db
python scripts/relabel_theology_sources.py $DB      # no-op if already relabelled
python scripts/import_heiser_content.py $DB
python scripts/import_bradley_content.py $DB
python scripts/ingest_reformed_evangelical.py $DB    # Owen, Stott, Lennox
python scripts/ingest_burnett_gage.py --theology-only $DB
python scripts/dedupe_theology_theme_links.py $DB   # no-op once clean

The relabel step must come first on a DB built before the descriptive labels: the data files now use those labels as source_work, so an unrelabelled DB would not match and would gain duplicate rows. The dedupe step removes theme links that earlier importer runs duplicated (their NULL reference slipped past the UNIQUE constraint), which inflated theme entry counts.

Ancient Near East Context

The 87 ANE entries are likewise original summary prose describing scholarly consensus on ANE background, with the standard reference works named as sources rather than quoted.

Vector Embeddings

Semantic search uses OpenAI's text-embedding-3-small model via sqlite-vec for the find_similar_passages tool.

Contributing

Contributions welcome. See CONTRIBUTING.md for setup, the test suite (run in CI on every pull request) and the rules every change keeps: small outputs, Scripture kept distinct from other sources, pinned and licensed data, and safe database upgrades. Especially welcome:

• Additional thematic cross-references
• Improved genre detection
• New translation support
• Better morphology parsing
• Documentation improvements

License

MIT License. See LICENSE.

Data from STEPBible is licensed under CC BY 4.0. OpenScriptorium LXX morphology and MT/LXX verse mappings (derived in part from STEPBible's TVTMS) are licensed under CC BY 4.0; the Rahlfs Greek text and Brenton's English translation and Greek text are public domain. The Perseus Josephus Greek and English TEI editions are licensed under CC BY-SA 4.0. The Onkelos Aramaic (Torat Emet) is public domain as declared by Sefaria; the Metsudah Chumash English of Onkelos is licensed under CC BY-NC 4.0 (non-commercial use only), and its licence is shown on every passage. The ETCBC/dss Dead Sea Scrolls transcription (Abegg, Bowley and Cook), the source of all four Qumran texts, is licensed under CC BY-NC 4.0 (non-commercial use only). The First1KGreek Philo Greek and English TEI editions are licensed under CC BY-SA 4.0; Cohn–Wendland's Greek and Yonge's English are public domain. Charles's 1 Enoch (via English Wikisource) and the KJV 2 Esdras (via eBible.org) are public domain; the Latin Wikisource transcription of the Clementine 4 Esdras is licensed under CC BY-SA 4.0. Mishnah passages carry each version's licence: Kulp's Mishnah Yomit English is CC BY, and the Vilna Mishnah is public domain. Data from BibleAquifer and ACAI is licensed under CC BY-SA 4.0. Torah Weave data (version 1.1) from Moshe Kline (chaver.com) is licensed under CC BY 4.0. BDB, Abbott-Smith, the Treasury of Scripture Knowledge, and the works of John Owen are public domain. The theology and ANE entries are original summary prose written for this project — see Data Sources.

Acknowledgments

• STEPBible for the freely available biblical data
• OpenScriptorium for Rahlfs LXX morphology and explicit MT/LXX verse mappings
• STEPBible.org / Tyndale House Cambridge for TVTMS, the versification data behind those mappings
• eBible.org for the public-domain Brenton Septuagint, English and Greek
• Perseus Digital Library for the licensed Josephus Greek and English editions
• Sefaria for the Torat Emet text of Onkelos and the Metsudah Chumash English, via Sefaria-Export-Archive
• ETCBC/dss and Martin G. Abegg Jr., James E. Bowley and Edward M. Cook for the Dead Sea Scrolls transcription
• Open Greek and Latin / First1KGreek for the licensed Philo Greek and English editions
• Wikisource contributors for R. H. Charles's 1 Enoch and the Latin Clementine Vulgate 4 Esdras
• eBible.org for the public-domain KJV with Apocrypha
• Sefaria for the Vilna Mishnah and Joshua Kulp's Mishnah Yomit English
• BibleAquifer for study notes, dictionary, translation notes, and key terms
• eliranwong for the unabridged BDB Hebrew lexicon
• translatable-exegetical-tools for the Abbott-Smith Greek Lexicon TEI XML edition
• Theographic for genealogy and event graph data
• Gordon Fee & Douglas Stuart for the hermeneutical framework in "How to Read the Bible for All Its Worth"
• The MCP community for the protocol and tooling