A production-ready MCP server that brings semantic search, multi-query expansion, and LLM-powered re-ranking across the full corpus of Ellen G. White writings into any AI agent or workflow.
Every query goes through a 5-step AI pipeline designed to find the most relevant passages — not just keyword matches.
include_answer: true,
gpt-5.3-chat produces a deep theological response with verbatim quotes, bilingual citations
(original English + translation into the user's language), and commentary.
507 Ellen G. White books, letters, and manuscripts are indexed. Here are the major works.
Plus hundreds of letters, manuscripts, periodical articles, and minor compilations — 507 sources total.
Use get_book_list tool to query the full index.
Read-only, idempotent tools exposed over the MCP protocol.
Universal search. Auto-detects page references and switches to exact metadata filter; otherwise runs the full 5-step AI pipeline.
"GC". Optional.{
"name": "egw_search",
"arguments": {
"query": "что Елена Уайт писала о молитве",
"top_k": 8,
"include_answer": true
}
}
Comprehensive topic search — returns all passages on a theme. Higher default top_k. Use when you want a full collection of quotes on a subject.
"prayer", "воспитание детей".Returns the full content of a specific page using exact Pinecone metadata filter. Fast and precise when you already know the page reference.
"DA". Optional.Streamable HTTP transport — the modern MCP standard. Works with Claude, n8n, Python, and any MCP client.
Add to claude_desktop_config.json:
{
"mcpServers": {
"egw": {
"type": "http",
"url": "https://egw.integrateflowsystems.com/mcp/"
}
}
}
In an MCP Client Tool node, set transport to Streamable HTTP:
https://egw.integrateflowsystems.com/mcp/
Use the official mcp library:
from mcp.client.streamable_http import streamablehttp_client async with streamablehttp_client( "https://egw.integrateflowsystems.com/mcp/" ) as (r, w, _): # use r/w session ...
Initialize a session with a POST to the endpoint:
POST /mcp/ HTTP/1.1
Host: egw.integrateflowsystems.com
Content-Type: application/json
Accept: application/json, text/event-stream
{"jsonrpc":"2.0","method":"initialize",
"id":1,"params":{"protocolVersion":"2025-03-26",
"clientInfo":{"name":"my-client","version":"1.0"},
"capabilities":{}}}
The engine understands natural language in any language — English, Russian, Spanish, German, and more.
Test the full pipeline live — query understanding, HyDE, multi-search, re-ranking, and answer generation. Sources appear instantly, answer streams in real time.