Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/exegia/corpora-py/llms.txt

Use this file to discover all available pages before exploring further.

When an AI assistant first connects to the cf-mcp server, it receives an instruction that outlines the recommended exploration sequence. Following this six-step workflow prevents wasted calls, avoids fetching unexpectedly large result sets, and ensures the agent builds up enough context about the corpus structure before querying it.
1

Understand Corpus Structure

Call describe_corpus() first. It returns the section hierarchy, total feature count, and every node type with its population count — giving the agent a complete map of what is in the corpus before any queries are run.
Section hierarchy: book > chapter > verse
Total features: 80

Node types:
Type                       Count
----------------------------------
word               426,584
verse               23,213
chapter              1,189
book                    39
The section hierarchy tells the agent how references like "Genesis 1:1" are structured, and the node type counts show the relative size of each level.
2

Browse Available Features

Call list_features() to see all annotation dimensions available for querying. Filter by node type to keep the list focused:
list_features(node_type="word", limit=30)
This returns feature names like lex, pos, gloss, gn, nu, st, vs, vt, sp, pdp, and others — the vocabulary of conditions the agent can use inside a search template.
3

Learn the Query Syntax

Call search_syntax_guide() for full inline documentation, or request a specific section to keep the context window lean:
search_syntax_guide(section="examples")
Available sections are nodes, relations, quantifiers, and examples. Skipping this step before crafting a complex template is the most common cause of malformed queries.
4

Check Result Scale

Before fetching any result tuples, estimate the size of the result set with return_type="count". This is a fast operation — the corpus engine runs the query but returns only the total:
search(template="word pos=verb", return_type="count")
# Returns: Total results: 72,471
Use this count to decide whether to paginate, narrow the template with additional conditions, or export to CSV instead.
5

Retrieve Paginated Results

Use return_type="results" to get the first page of node tuples. The response includes a cursor_id that can be passed to search_continue() to walk through the full result set page by page:
search(template="word pos=verb", return_type="results", limit=20)
# Returns results 1–20 and: cursor_id: abc-123...

search_continue(cursor_id="abc-123", limit=20)
# Returns results 21–40
Each call to search_continue() advances the cursor and reports how many results remain.
6

Read Passage Text

Use get_passages() to retrieve the rendered text for any section references returned by the search. Pass an explicit fmt to select a specific text encoding — omitting it uses the corpus default:
get_passages(references=["Genesis 1:1", "John 3:16"], fmt="text-orig-full")
References follow the "Book Chapter:Verse" pattern. The tool resolves each reference to a corpus node and returns the formatted text.

​
Quick Reference

describe_corpus()      →  understand node types and section hierarchy
list_features()        →  see annotation dimensions available for search
search_syntax_guide()  →  learn the query language
search(..., "count")   →  estimate result size before fetching
search(..., "results") →  get paginated node tuples with a cursor_id
search_continue()      →  page through large result sets
get_passages()         →  read the actual rendered text
The search() tool supports four return_type values, each suited to a different stage of the workflow:
return_typeWhat it returns
resultsNode tuples with section refs and a cursor_id for pagination
countTotal result count as a formatted number
statisticsNode type breakdown of all matched nodes
passagesRendered text for the first N matches

​
Cursor TTL

Cursors created by search() expire after 5 minutes of inactivity. The server automatically purges expired cursors before each new search() or search_continue() call. If a cursor has expired, search_continue() returns "Cursor expired or not found. Run search() again." — simply re-run the original search() to get a fresh cursor.
Always check result count with return_type="count" before requesting return_type="results" on an unfamiliar template. Large corpora can return millions of results, and fetching all of them page by page may be impractical.

Build docs developers (and LLMs) love

Get started for freeTalk to us