Happy slice of mcp pizzaMCP.pizza

MODULAR-RAG-MCP-SERVER

MCP.Pizza Chef: jerry-ai-dev

This tool builds a searchable knowledge base out of your own PDFs and images, then lets Claude Desktop, Cursor, or VS Code pull relevant passages back into a conversation to answer questions grounded in your material. It combines keyword and meaning-based search so both exact terms and paraphrased questions find the right passage, then double-checks results before handing them to your assistant. Setting it up requires a key for the AI model it runs on and some technical configuration, so it is best suited to people comfortable installing developer tools.

Data
Files/PDF

Use This MCP server To

Ask questions and get answers pulled from my own documents Find a passage even when I don't remember the exact wording Get a quick summary of what's inside a document collection Search image content that's been described in plain text See which document an answer came from before trusting it Browse the different document collections I've set up

README

Modular RAG MCP Server

一个可插拔、可观测的模块化 RAG（检索增强生成）服务框架，通过 MCP（Model Context Protocol）协议对外暴露工具接口，支持 Copilot / Claude 等 AI 助手直接调用。同时也是一份专为大模型相关岗位学习与面试求职设计的实战项目与配套教学资源。

📖 目录

🏗️ 项目概述

这个项目是什么

本项目将 RAG 面试中最常见的核心环节——检索（Hybrid Search + Rerank）、多模态视觉处理（Image Captioning）、RAG 评估（Ragas + Custom）、生成（LLM Response）——以及当下热门的应用协议 MCP（Model Context Protocol） 串联为一个完整的、可运行的工程项目。

项目的一大亮点是极易适配到你自己的业务中。得益于全链路可插拔架构，你可以快速将它结合到自己已有的项目里，无论你的背景和需求如何，都能找到适合自己的使用方式。具体的使用策略会在后文 谁适合用这个项目 & 怎么用 中详细展开。

不只是项目，更是一整套思路

比这个项目本身更有价值的，是它背后蕴含的一整套工程化思路：

  • 如何编写 DEV_SPEC（开发规格文档）来驱动开发
  • 如何用 Skill 基于 Spec 自动完成代码编写
  • 如何用 Skill 进行自动化测试、打包、环境配置
  • 如何基于可插拔架构进行扩展（比如扩展到 Agent）

学会了思路，你可以自己做全新的项目和扩展。以上每一步的具体做法、设计思路，在笔记中都有对应的视频讲解，建议配合观看。

MODULAR-RAG-MCP-SERVER FAQ

Can I use this to search my own PDFs and get answers with sources?
Yes — you load your documents in first, and once they're ready, your assistant can search them and point back to which document an answer came from.
Can I use this to search pictures for what they show, not just text?
Yes — the tool writes a plain-language description of each image as it loads your documents, so searching for what's shown in a picture can also turn up the right file.
Which apps can I connect this to?
It's built to work with Claude Desktop, Cursor, and VS Code's Copilot chat, using the standard connection method those apps support.
Do I need an account or API key to use it?
Yes — you'll need an API key for the AI model that powers the search and answers, such as OpenAI, Claude, or Gemini, and you'll set up your own document storage as well.
How hard is it to set up?
This one takes developer-level setup — installing it from source code and configuring several pieces, including document storage, search settings, and your API key, rather than a quick one-click install.
Will it slow down or change my regular chats?
No — it only runs when your assistant specifically asks it to search your documents, so everyday conversations are unaffected.
Can it handle documents that aren't in English?
The project's own instructions are written in Chinese, but the search and answering approach isn't tied to one language, so it can generally work with documents in other languages too, though it's worth double-checking results.
Is there a way to see how well the search results are performing?
Yes — it includes a built-in dashboard where you can review past searches, see which documents were pulled up, and track answer quality over time.