MCP Servers

A collection of Model Context Protocol servers, templates, tools and more.

R
Rag MCP
by @v833
GitHub Repo(1 stars)

MCP server by v833

Created 3/29/2026
Updated about 7 hours ago
README
View Source
Repository documentation and setup instructions

My RAG

A full-stack document processing and retrieval MVP for building local-first RAG workflows.

一个面向文档处理、语义检索与回答生成的全流程 RAG 原型项目。

Frontend Backend Build Status

Overview | 项目简介

English

My RAG is a visual, end-to-end RAG playground that helps you go from raw documents to searchable knowledge and citation-aware answers. It includes upload, parsing, chunking, embedding, indexing, semantic search, and answer generation in one interface.

中文

My RAG 是一个可视化的端到端 RAG 演示项目，覆盖从原始文档上传，到结构化解析、文本分块、向量化、索引、语义搜索，再到带引用回答生成的完整流程。项目默认采用本地 JSON 持久化每一步结果，方便调试、教学演示和二次开发。

Highlights | 特性亮点

  • End-to-end RAG pipeline with 7 linked UI pages
  • Local-first storage with persistent JSON outputs for each stage
  • Support for pdf / txt / md / docx
  • Parse modes for full text, per-page, and heading-aware structure extraction
  • Chunk strategies for heading_aware, fixed_length, and per_page
  • OpenAI and HuggingFace embedding/generation support
  • Fallback local embedding and answer generation when API keys are missing
  • Milvus/Pinecone-style indexing entrypoints backed by local JSON collections
  • FastAPI backend plus Vue 3 frontend for quick iteration

Screenshots | 项目截图

当前仓库还没有真实页面截图，README 已先放入占位图，后续你可以直接替换这些文件：

  • docs/images/dashboard-overview.svg
  • docs/images/pipeline-detail.svg
  • docs/images/search-generation.svg

Dashboard Overview

Pipeline Detail

Search And Generation

Workflow | 处理流程

flowchart LR
    A["Upload"] --> B["Load"]
    B --> C["Parse"]
    C --> D["Chunk"]
    D --> E["Embedding"]
    E --> F["Indexing"]
    F --> G["Search"]
    G --> H["Generation"]

对应的前端页面为：

  • LoadFile
  • ParseFile
  • ChunkFile
  • EmbeddingFile
  • Indexing
  • Search
  • Generation

Architecture | 架构图

flowchart TB
    subgraph UI["Frontend · Vue 3 + Vite"]
        P1["LoadFile"]
        P2["ParseFile"]
        P3["ChunkFile"]
        P4["EmbeddingFile"]
        P5["Indexing"]
        P6["Search"]
        P7["Generation"]
    end

    subgraph API["Backend · FastAPI"]
        R1["/api/load"]
        R2["/api/parse"]
        R3["/api/chunk"]
        R4["/api/embedding"]
        R5["/api/indexing"]
        R6["/api/search"]
        R7["/api/generation"]
    end

    subgraph Service["Core Services"]
        S1["loading_service"]
        S2["parse_service"]
        S3["chunk_service"]
        S4["embedding_service"]
        S5["indexing_service"]
        S6["search_service"]
        S7["generation_service"]
        S8["model_gateway"]
    end

    subgraph Data["Storage"]
        D1["backend/data/uploads"]
        D2["backend/data/*_results"]
        D3["backend/data/index_collections"]
    end

    P1 --> R1 --> S1 --> D1
    P2 --> R2 --> S2 --> D2
    P3 --> R3 --> S3 --> D2
    P4 --> R4 --> S4 --> S8
    P5 --> R5 --> S5 --> D3
    P6 --> R6 --> S6 --> S8
    P7 --> R7 --> S7 --> S8
    S4 --> D2
    S5 --> D2
    S6 --> D2
    S7 --> D2

Tech Stack | 技术栈

Frontend

  • Vue 3
  • TypeScript
  • Vite
  • Vue Router
  • Tailwind CSS

Backend

  • FastAPI
  • Pydantic
  • Uvicorn
  • httpx

Document Processing

  • PyMuPDF
  • pypdf
  • unstructured
  • python-docx

Testing

  • pytest

Repository Structure | 仓库结构

.
├── frontend/                     # Vue 3 前端
│   ├── src/pages/                # 工作流页面
│   ├── src/api/                  # 前端 API 封装
│   ├── src/layouts/              # 布局组件
│   └── src/components/           # 通用组件
├── backend/                      # FastAPI 后端
│   ├── app/api/                  # 路由层
│   ├── app/services/             # 核心业务逻辑
│   ├── app/models/               # Pydantic 模型
│   ├── app/core/                 # 配置与常量
│   ├── app/utils/                # 文件存储等工具
│   ├── data/                     # 本地输入、输出与索引数据
│   └── tests/                    # 测试
├── docs/images/                  # README 配图
└── README.md

后端会自动创建并使用这些数据目录：

backend/data/uploads/
backend/data/load_results/
backend/data/parse_results/
backend/data/chunk_results/
backend/data/embedding_results/
backend/data/index_results/
backend/data/search_results/
backend/data/generation_results/
backend/data/index_collections/

Quick Start | 快速开始

1. Start Backend

cd backend
python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
uvicorn app.main:app --reload

Backend default:

http://127.0.0.1:8000

Health check:

curl http://127.0.0.1:8000/api/health

2. Start Frontend

cd frontend
npm install
npm run dev

Frontend default:

http://127.0.0.1:5173

If your backend runs elsewhere:

VITE_API_BASE_URL=http://127.0.0.1:8000

Environment Variables | 环境变量

The backend reads backend/.env. The project can still run most demo flows without API keys because it includes local fallback logic.

OPENAI_API_KEY=
OPENAI_EMBEDDING_MODEL=text-embedding-3-small

HUGGINGFACE_API_KEY=
HUGGINGFACE_EMBEDDING_MODEL=sentence-transformers/all-MiniLM-L6-v2

BACKEND_CORS_ORIGINS=http://127.0.0.1:5173,http://localhost:5173

Suggested backend/.env:

OPENAI_API_KEY=your_key_here
OPENAI_EMBEDDING_MODEL=text-embedding-3-small
HUGGINGFACE_API_KEY=
HUGGINGFACE_EMBEDDING_MODEL=sentence-transformers/all-MiniLM-L6-v2
BACKEND_CORS_ORIGINS=http://127.0.0.1:5173,http://localhost:5173

Pages | 页面说明

LoadFile

  • Upload raw files
  • Choose auto / pymupdf / pypdf / unstructured
  • Produce standardized LoadResult JSON with full text and page previews

ParseFile

  • Transform LoadResult into structured ParseResult
  • Support by_heading / per_page / full_text
  • Generate reusable blocks for downstream chunking

ChunkFile

  • Convert ParseResult into ChunkResult
  • Support heading_aware / fixed_length / per_page
  • Preserve page ranges, heading paths, and source block references

EmbeddingFile

  • Prefer chunk-based embeddings
  • Keep compatibility with legacy page-level embedding from LoadResult
  • Show vector previews and lightweight 2D projections

Indexing

  • Write chunk embeddings into collections
  • Expose milvus / pinecone provider names
  • Current implementation uses local JSON collections instead of real vector databases

Search

  • Run semantic retrieval on saved collections
  • Support top_k, file filter, and page-range filter
  • Persist search results for reuse in generation

Generation

  • Generate answers from saved search results
  • Support OpenAI and HuggingFace
  • Return answer text plus citations

API Summary | API 概览

  • POST /api/load/upload
  • POST /api/load/process
  • GET /api/load/uploads
  • GET /api/load/files
  • POST /api/parse/process
  • GET /api/parse/files
  • POST /api/chunk/process
  • GET /api/chunk/files
  • POST /api/embedding/run
  • GET /api/embedding/tasks
  • POST /api/indexing/run
  • GET /api/indexing/tasks
  • GET /api/indexing/collections
  • POST /api/search/run
  • GET /api/search/results
  • POST /api/generation/run
  • GET /api/generation/results

Testing | 测试

Run backend tests:

cd backend
pytest

Current coverage includes:

  • filename sanitization
  • JSON persistence helpers
  • heading detection
  • heading-aware chunk grouping
  • fallback embedding stability

Current Scope | 当前范围

  • milvus and pinecone are currently local JSON-backed mock providers
  • Embedding and generation can fall back to local logic when external credentials are not configured
  • This repository is best suited for demos, teaching, prototyping, and secondary development

Roadmap Ideas | 可扩展方向

  • Connect to real Milvus, Pinecone, or pgvector
  • Add reranking and richer retrieval strategies
  • Support streaming answers and conversation memory
  • Add OCR and more file format pipelines
  • Add auth, dataset management, and multi-user capabilities

License

You can add an open-source license such as MIT before publishing publicly.

Quick Setup
Installation guide for this server

Install Package (if required)

uvx rag-mcp

Cursor configuration (mcp.json)

{ "mcpServers": { "v833-rag-mcp": { "command": "uvx", "args": [ "rag-mcp" ] } } }