Skip to main content
The API hosts public domain texts of the world’s religions and traditions. Each one is numbered the way people cite it and linked to the Urantia paragraphs closest in meaning. kind is scripture for a text that its tradition reads as sacred, and classic for a story, history, or teaching of a tradition that is not read as scripture. Seven of the texts belong to religions with a section in Paper 131, “The World’s Religions”: 131:1 (the two Cynic texts), 131:3 (Dhammapada), 131:4 (Gita), 131:7 (Shinto oracles), 131:8 (Tao Te Ching), and 131:9 (Analects). Paper 131 has no section headed Islam or Sikhism. The Upanishads are three of the twelve in Sacred Books of the East, volume 1: the Khândogya, the Kena (Talavakâra), and the Îsâ. Each title links to the page for that text. For the Bible (World English Bible), use the Bible endpoints, or /works below, which serves the Bible and these texts together.

One list for every work

Each result has the same shape: the work (id, title, tradition), the reference, the text, a url to the public page that holds the passage, and the nearest Urantia paragraphs. The MCP server has the same calls as the works.* tools.

List the texts

Each entry names its translator, year, source, and license, and notes says how the text is numbered or selected:
GET /scriptures/{corpus} returns one text with its chapters (books, for the Analects) and the first and last reference of each. The corpus can be the slug, the id, or the ref prefix: bhagavad-gita, bhagavad-gita-besant-1922, and bg are the same.

Get passages

References:
  • Bhagavad Gita: chapter and verse, BG 2.47
  • Koran: surah and verse, Quran 2.255, in the standard count of 6,236 verses
  • Analects: book and chapter, Analects 15.23
  • Diogenes Laertius: book and section, DL 6.20. Each section also names the life it belongs to, such as Diogenes
  • Tao Te Ching: chapter and paragraph, TTC 25.1. Ask for 25 to get the whole chapter
  • Epictetus: book, chapter, and section, Epictetus 3.22.45. Oldfather numbers every fifth section, so each passage covers up to five sections, and 3.22.47 returns Epictetus 3.22.45-49. Ask for 3.22 to get the whole chapter
  • Dhammapada: the verses are numbered straight through the book, so Dhp 183 is verse 183, not chapter 183. Muller combines a few pairs of verses, such as Dhp 58-59. Asking for either number returns the pair
  • Japji: Japji 0 is the opening (the Mul Mantar), Japji 1 to Japji 38 are the pauris, and Japji 39 is the closing slok
  • Shinto oracles: Oracle 1 to Oracle 15. Aston does not number them, so the numbers follow his order, and each oracle carries its shrine as its title
  • Book of Mormon: book, chapter, and verse by name, 1 Nephi 3:7 or Alma 32:21
  • Rig Veda: book, hymn, and verse, RV 10.129.1
  • Book of the Dead: chapter (spell) and paragraph, BD 125.3. The paragraph numbers are ours
  • Hesiod: poem and line, Theogony 116 or Works and Days 293. A passage covers a range of lines, such as Works and Days 293-319
  • Homeric Hymns: hymn and line, HH 2.1
  • Poetic Edda: poem and stanza by name, Voluspo 3
  • Aztec hymns: hymn and verse, Aztec Hymn 1.3
  • Chuang Tzu: chapter and paragraph, Zhuangzi 2.5. The paragraph numbers are ours
  • Prose Edda: part, chapter, and paragraph, Gylfaginning 1.1. The parts are Prologue, Gylfaginning, and Skáldskaparmál. The paragraph numbers are ours
  • Kojiki: section and paragraph, Kojiki 1.1. The paragraph numbers are ours
  • Upanishads: Khandogya 6.8 (prapathaka and khanda; each passage is one whole khanda, with its verse numbers in the text), Kena 1.3 (khanda and verse), and Isa 8 (verse)
  • Shu King: part, book, and paragraph, Shu 5.4.6. Legge prints his own numbers in only some books, so the paragraph numbers are ours, in order through each book
refLevels on each text says how many numbers a full reference has. The prefix is optional, and 2:47 works the same as 2.47.

Urantia paragraphs for a passage

Returns the passage, the chunk that holds it, and the 10 closest Urantia paragraphs. A chunk is a group of short passages of about one Urantia paragraph in length. A Tao Te Ching chunk is one whole chapter.

Passages for a Urantia paragraph

Add scriptureParallels to include on any single-paragraph endpoint:
The response gains a scriptureParallels array: the 3 closest chunks from each text, grouped by text in the order of the list. Each entry carries its corpus (title, religion, translator, year), the chunk reference and text, and the similarity and rank. It also works with format=rag.

Search by meaning

Searches every text, or one with "corpus": "bhagavad-gita". Each result carries its top Urantia paragraphs; set urantiaParallelLimit from 0 to 10 (default 3).

Patterns across the texts

Four lists rank the Urantia paragraphs by how they relate to all the texts at once (the texts above and the Bible):
Add scriptureScores to include on a paragraph to get its own numbers: how many texts it is close in, its lean, and its pairs. How the scores work:
  • For each text, a paragraph’s closest passage is ranked against every other paragraph’s, after an adjustment for paragraph length. Longer paragraphs score higher, so without the adjustment the lists favor them.
  • “Close in a text” means the paragraph is in that text’s top 10%.
  • The texts often move together, because a paragraph that sounds devotional is close to many of them. A count of texts is not a count of independent votes.
  • A lean and a pair count only when two embedding models, text-embedding-3-large and text-embedding-3-small, agree. This check removed more than half of the candidate pairs.
  • Paragraphs under 200 characters, such as headings, are left out by default (minLength).
  • Subjects the scriptures do not discuss, such as geology, rank as far from every text. Filter far by partId or paperId to compare paragraphs on one subject.

In Spanish

The texts are English. With lang=es, /works, /scriptures, and /bible/books give the titles, traditions, summaries, labels, chapter titles, and book names in Spanish, as an AI translation. The passages stay English, and each work says so with textLanguage: "eng". See Translations.

What a parallel is

A parallel is a pair: one Urantia paragraph and one passage of another text that are close in meaning. The API finds them with one method, applied the same way to all 16,220 paragraphs and to every passage of the Bible and the 21 other texts. So every paragraph has its parallels, and every result can be checked and reproduced. Because the method reads meaning and not only words, it finds passages that treat the same subject in other words, across traditions and translations, where a search by words finds nothing. A parallel shows that two passages are close in meaning. It does not claim that one text came from the other.

How the API computes them

  1. Cut each text into passages. A passage (a chunk) is about as long as one Urantia paragraph, and it never crosses a chapter, a book, or a titled part. The Tao Te Ching is cut by whole chapters.
  2. Turn each passage into an embedding. OpenAI’s text-embedding-3-large model turns each chunk and each Urantia paragraph into a list of 3,072 numbers that describes its meaning. Passages with close meanings get lists that point in close directions, also when they share no words.
  3. Compare every pair. The API compares each paragraph with each chunk of each text. The comparison is exact, not an estimate.
  4. Keep the nearest. For each paragraph it keeps the 3 nearest passages of each text (?include=scriptureParallels) and the 10 nearest Bible passages (?include=bibleParallels). For each passage it keeps the 10 nearest paragraphs.
The API runs these steps again when a text is added or changed, so the parallels always match the texts that it serves.

How well it works

As a check, we asked whether each text finds its own religion’s section of Paper 131, which gathers teachings of those religions. For the Dhammapada, the Gita, the Tao Te Ching, and the Analects, 69% to 93% of the chunks have that section in their top 10 Urantia paragraphs. 60% of the Shinto oracles have 131:7 in their top 10. By chance it would be well under 1%. For a study that compares Paper 131 with earlier anthologies of world scripture, see Matthew Block’s source studies.

What the number means

similarity is the cosine similarity of the two embeddings, from 0 to 1. Higher is closer.
  • From about 0.46 up, a passage of another text is most often close in meaning. From 0.40 to 0.46 most are. Below 0.40 the match is often only in words or style, so UrantiaHub, for example, shows passages from 0.40 up.
  • Compare numbers inside one text. Urantia paragraphs score 0.58 to 0.73 among themselves, so in a list mixed with other texts they come first.
  • Every row also has rank, source ("semantic"), and embeddingModel, so you can filter by strength and check each result.

Computed and curated parallels

A list made by hand holds the links that one person picked, most often word-level allusions, for part of the text. The computed parallels cover every paragraph and twenty-two texts with one method, and they find links of meaning that share no words. Like any reader, the model can miss an allusion, so the similarity number is there to show how strong each link is.

Sources

The text keeps the translator’s own words and spelling, including the parentheses that Muller, Legge, and Pickthall use for words they added. Page headers, footnotes, and the Chinese and Sanskrit lines are not included. Aston’s own summaries of oracles he does not quote are not included either.