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.

A single cf-mcp process can hold multiple corpora in memory at the same time. You can load them at startup with repeated --corpus flags, give each one a short name, and then target any of them in individual tool calls without restarting the server or changing the global default.

​
CLI: Multiple —corpus Flags

Repeat --corpus for each dataset you want to load. Pair each path with an optional --name flag to give it a human-readable identifier: 
uv run cf-mcp \
  --corpus ~/.exegia/datasets/bibles/BHSA --name BHSA \
  --corpus ~/.exegia/datasets/bibles/GNT  --name GNT
--name flags are matched positionally to --corpus flags — the first --name applies to the first --corpus, and so on. If a --name is omitted for a given path, the name defaults to the directory’s base name.

​
—features Flag

To speed up startup, pass --features with a list of feature names to load only the annotations you need. This is particularly useful for large corpora with dozens of features when your workflow only requires a subset: 
uv run cf-mcp \
  --corpus ~/.exegia/datasets/bibles/BHSA \
  --features lex pos gloss
Load a small subset of features with --features to speed up startup when you only need specific annotations.

​
Programmatic Multi-Corpus

When embedding the server in Python, call corpus_manager.load() once per corpus before calling mcp.run(). The first corpus loaded automatically becomes the default: 
from exegia.mcp import corpus_manager, mcp

corpus_manager.load("~/.exegia/datasets/bibles/BHSA", name="BHSA")
corpus_manager.load("~/.exegia/datasets/bibles/GNT",  name="GNT")

# First loaded corpus is the default
print(corpus_manager.current)       # "BHSA"
print(corpus_manager.list_corpora()) # ["BHSA", "GNT"]

# Switch the active corpus
corpus_manager.select("GNT")

mcp.run(transport="stdio")

​
Using the corpus Parameter in Tools

Every one of the 11 MCP tools accepts an optional corpus parameter. Passing it directs that single call to the named corpus without changing the global default — useful when an AI agent needs to compare results across datasets in the same conversation: 
search(template="word pos=verb", corpus="GNT")
get_passages(references=["John 3:16"], corpus="GNT")
describe_corpus(corpus="BHSA")
If corpus is omitted, the tool uses whichever corpus is currently set as the default.

​
list_corpora Tool Output

The list_corpora tool shows all loaded corpora and marks the active one with an asterisk: 
Loaded corpora (* = current):
  * BHSA
    GNT

​
CorpusManager API Reference

MethodSignatureReturns
loadload(path, name?, features?)str — the corpus name
get_apiget_api(name?)Context-Fabric TF API object
list_corporalist_corpora()list[str]
selectselect(name)None
unloadunload(name)None
get_api() with no argument returns the API for the currently active corpus. After unload(), the active corpus is automatically advanced to the next loaded corpus, or set to None if the last corpus was removed.

Build docs developers (and LLMs) love

Get started for freeTalk to us