# Search and AI FAQ

What AI does, what your AI receives, and how to check the source.

## Where does AI work in this library?

Qwen3 embeddings find passages by meaning. Full-text search finds matching words. Reciprocal rank fusion combines the two rankings, then the service loads the stored source text and citations. This is hybrid search: the retrieval part of RAG.

We run the embedding model and maintain the index. You do not need to host a retrieval model, build an index, or provide an API key for retrieval.

## What changes when I connect my own AI through MCP?

Website search returns passages for you to read and interpret. Through MCP, your AI can retrieve the same source records and use them as context. When it generates an answer from those records, the combined workflow is retrieval-augmented generation (RAG).

MCP provides access to retrieval. It does not generate an answer by itself. Your AI provider supplies generation and may require its own account or paid plan.

## Does my AI receive actual text or generated summaries?

It receives the full stored passage text, its citation, source-type label, and source links. Meaning matches include an excerpt copied directly from that passage. Our service does not generate or rewrite the source text.

The records distinguish the historical text from editorial material where classified and preserve uncertainty where it is not. Your AI can write its own explanation, but that explanation is separate from the passages we supply.

## How does this differ from my AI using general web search?

This library searches a defined 38-volume collection of historical English texts. General web search can return original texts, commentary, summaries, and quotation sites. Both can be useful, but the agent must assess what each source represents.

Our results include structured citations and source-type labels. The tools also expose full passages, notes, neighboring sections, and printed-page evidence. With web search, the agent must find and assemble that context from what each website makes available.

The benefit comes from the source collection and retrieval tools, not from MCP itself. They give your AI a direct route to inspectable evidence. General web search can also find scholarship and material outside this collection.

## Is MCP more accurate than web search?

We have not measured an accuracy advantage in a direct comparison with general web search. A careful web-search agent could find the same passage and answer equally well.

The design aims to make source grounding more consistent by supplying passages, citations, and context through defined tools. It can still retrieve an irrelevant passage, and your AI can misinterpret a correct one. Check the evidence behind its answer.

## Can I check the full text and actual scanned page?

Yes. Search results include source transcription and scan links. The get_passage tool returns the full passage, notes, Bible references, neighboring passage IDs, and page-level scan evidence. The get_work tool lists sections in source order so your AI can read more of the work.

On the website, the passage reader places text beside the scan on desktop and offers Read/Scan tabs on mobile. Scan availability and comparison coverage vary. A scan link alone does not mean that its alignment or every transcribed character has been verified.

## Do I need to choose a search mode?

No. The website, API, and MCP use auto mode. Questions and concepts use hybrid search, which combines meaning and words. Quoted phrases, Bible references, passage IDs, recognized names or Greek and Latin terms, and boolean or wildcard syntax use word search. Bible references also use the indexed Scripture lookup.

Each response includes mode_used and mode_reason. The website shows a quiet label and a link to search exact words instead. Agents should omit mode unless you explicitly ask for exact words. If meaning search is unavailable or reaches its monthly allowance, search continues with words.

## Explore further

- [How the AI works and a sample MCP request](/connect#how-ai-works)
- [Connect your AI](/connect)
- [Source verification coverage](/verification)
- [Related example: Microsoft Learn MCP FAQ](https://learn.microsoft.com/en-us/training/support/mcp-faq)