From 740d0c3c64ef91e940b36b9cffacaae9e0659c3c Mon Sep 17 00:00:00 2001 From: Masha_Rudenko Date: Mon, 10 Aug 2026 12:41:48 +0300 Subject: [PATCH] [update] add Search/Inference and example to Suite MCP guide - how-it-works section lacked the Search/Inference workflow split that every other product guide explains, and had passive-voice steps with no named actor - added a concrete walkthrough (Layout + Calendar + Grid prompt), framed as one example among the many Suite components covered --- docs/guides/mcp-server.md | 12 ++++++------ 1 file changed, 6 insertions(+), 6 deletions(-) diff --git a/docs/guides/mcp-server.md b/docs/guides/mcp-server.md index 102f78a8..d9e9e775 100644 --- a/docs/guides/mcp-server.md +++ b/docs/guides/mcp-server.md @@ -39,13 +39,13 @@ The MCP server indexes the full DHTMLX Suite documentation across all components The DHTMLX MCP server uses a Retrieval-Augmented Generation (RAG) pipeline combined with the Model Context Protocol (MCP) to provide AI assistants with up-to-date documentation. Before any of that, the assistant first figures out which part of a request actually needs a documentation lookup and handles the rest from its own knowledge. -At a high level: +Here's what that looks like for the prompt *"I want to create a layout with a calendar in one cell, and a grid in another,"* one example among the many Suite components this same mechanism covers: -1. The assistant sends the part of the query that needs documentation through MCP. -2. The server determines which product documentation is relevant. -3. Documentation content is retrieved from a vector index. -4. The retrieved context is sent back to the assistant. -5. The assistant combines that context with the part of the request it already handled on its own to generate a response. +1. The assistant sends the query through MCP. +2. The server determines that it touches the Layout, Calendar, and Grid documentation. +3. Since the answer requires generated code, the server routes the query to *Search*, one of two workflows; a narrower factual question would route to *Inference* instead, which reads the same pages and answers directly. +4. *Search* retrieves the matching pages from a vector index built on the current documentation and sends them back to the assistant as context. +5. The assistant configures the layout, calendar, and grid together using that context. This approach allows AI tools to generate answers based on current documentation.