Skip to content
Conduit

mcp server

Mindage

Mind Map Driven Knowledge Graph and Article Generation Platform

RangerCao/Mindagecommunitylistedunreviewed

Set up

Runs on your computer. Pick your agent and operating system.

Before you start

  • uv (`curl -LsSf https://astral.sh/uv/install.sh | sh`)
~/Library/Application Support/Claude/claude_desktop_config.json
{
  "mcpServers": {
    "mindage": {
      "command": "uvx",
      "args": [
        "mindage"
      ]
    }
  }
}

same format Cursor, GitHub Copilot CLI, Junie, JetBrains AI Assistant, Augment (Auggie CLI), Factory Droid, Kiro, Devin CLI +12

Values in angle brackets are placeholders. Restart Claude Desktop after editing its config.Claude Desktop docs

README

synced Oct 22, 2026

Mindage

Mind Map Driven Knowledge Graph & Article Generation Platform

思维导图驱动的知识图谱与文章生成平台


Mindage is an interactive knowledge platform built on top of LightRAG. It transforms traditional retrieval-augmented generation into an exploratory experience — navigate knowledge through mind maps, generate structured articles from topic trees, and extend capabilities through MCP tool integration.

✨ Key Features

  • Mind Map Exploration — Canvas-based interactive mind maps powered by LLM. Expand topics, generate per-node content, and produce full structured articles from a single topic tree.
  • Knowledge Graph Visualization — Interactive graph visualization of entity relationships extracted from your documents, powered by Sigma.js.
  • Multi-Mode Retrieval — Five query modes (local, global, hybrid, mix, naive) to balance precision and breadth for different query types.
  • MCP Tool Integration — Connect external tools via Model Context Protocol (stdio / SSE / streamable HTTP) and invoke them during conversations.
  • PageIndex Tree Index — Hierarchical document tree index for structured navigation of processed documents.
  • Multi-User RBAC — JWT-based authentication with admin / user / guest roles and self-service password management.
  • Multi-Workspace — Create isolated knowledge bases with separate document sets and knowledge graphs.
  • 11 Languages — Full internationalization via i18next including Chinese, English, Japanese, Korean, French, German, Spanish, and more.
  • Observability — Prometheus metrics, request tracing, and structured JSON logging for production monitoring.

📸 Screenshots

🚀 Quick Start

Prerequisites

  • Python 3.10+
  • Node.js 18+ & Bun
  • uv (recommended for Python package management)

Install from Source

git clone https://github.com/RangerCao/LightRAG.git
cd LightRAG

# Set up Python environment
uv sync --extra api
source .venv/bin/activate        # Linux/macOS
# .venv\Scripts\activate         # Windows

# Build WebUI
cd lightrag_webui
bun install --frozen-lockfile
bun run build
cd ..

# Configure environment
cp env.example .env
# Edit .env with your LLM and embedding API keys

# Start the server
lightrag-server

Docker Deployment

git clone https://github.com/RangerCao/LightRAG.git
cd LightRAG
cp env.example .env
# Edit .env with your configurations
docker compose -f docker-compose-full.yml up -d

The full Docker stack includes: Mindage server, vLLM embedding/reranking services, PostgreSQL (pgvector), Neo4j graph database, and Milvus vector database.

Access Mindage

Once running, open your browser:

  • WebUI: http://localhost:9621/webui/
  • API Docs: http://localhost:9621/docs
  • Health Check: http://localhost:9621/health

⚙️ Configuration

Mindage is configured via environment variables in the .env file. Key settings:

VariableDescriptionDefault
LLM_BINDINGLLM provider (openai / ollama / azure)openai
LLM_MODELLLM model namegpt-4o-mini
EMBEDDING_BINDINGEmbedding provideropenai
EMBEDDING_MODELEmbedding model nametext-embedding-3-small
SUMMARY_LANGUAGEOutput language for entities/relationsEnglish
ENABLE_PAGEINDEXEnable tree index featurefalse
MAX_PARALLEL_INSERTMax documents processed in parallel3

See env.example for the complete configuration reference.

🏗️ Architecture

Mindage
├── lightrag/              # Core Python package
│   ├── api/               # FastAPI server + REST endpoints
│   │   ├── routers/       # Document, query, graph, explore, MCP routes
│   │   └── lightrag_server.py
│   ├── kg/                # Storage backends (Neo4j, PostgreSQL, Milvus, etc.)
│   ├── llm/               # LLM providers (OpenAI, Ollama, Azure, etc.)
│   ├── parser/            # Document parsers (native, MinerU, Docling)
│   ├── chunker/           # Chunking strategies
│   ├── pipeline.py        # Document ingestion pipeline
│   └── operate.py         # Extraction & retrieval operations
├── lightrag_webui/        # React 19 + TypeScript frontend
│   └── src/
│       ├── components/    # UI components
│       ├── views/         # Page views (Chat, Knowledge, Graph, MindMap...)
│       └── i18n/          # 11 language translations
└── docker-compose-full.yml

🙏 Acknowledgments

Mindage is built on top of LightRAG by HKUDS. We gratefully acknowledge the original project's foundational contributions to graph-based retrieval-augmented generation.

If you use Mindage in your research, please cite the original LightRAG paper:

@article{guo2024lightrag,
  title={LightRAG: Simple and Fast Retrieval-Augmented Generation},
  author={Zirui Guo and Lianghao Xia and Yanhua Yu and Tu Ao and Chao Huang},
  year={2024},
  eprint={2410.05779},
  archivePrefix={arXiv},
  primaryClass={cs.IR}
}

Mindage 是一个基于 LightRAG 构建的交互式知识平台。它将传统的检索增强生成转化为一种探索式体验——通过思维导图导航知识、从主题树生成结构化文章,并通过 MCP 工具集成扩展能力。

✨ 核心功能

  • 思维导图探索 — 基于 Canvas 的交互式思维导图,由 LLM 驱动。展开主题、生成节点内容、从单一主题树生成完整的结构化文章。
  • 知识图谱可视化 — 基于 Sigma.js 的交互式图谱,展示从文档中提取的实体关系网络。
  • 多模式检索 — 五种查询模式(局部、全局、混合、融合、朴素),在不同查询场景下平衡精确度与广度。
  • MCP 工具集成 — 通过 Model Context Protocol(stdio / SSE / streamable HTTP)连接外部工具,在对话中直接调用。
  • PageIndex 树形索引 — 层级化文档树索引,支持结构化浏览已处理文档。
  • 多用户权限管理 — 基于 JWT 的认证系统,支持管理员 / 普通用户 / 访客三种角色,支持自助修改密码。
  • 多工作空间 — 创建隔离的知识库,每个空间拥有独立的文档集和知识图谱。
  • 11 种语言 — 基于 i18next 的完整国际化支持,包括中文、英文、日文、韩文、法语、德语、西班牙语等。
  • 可观测性 — Prometheus 指标、请求追踪和结构化 JSON 日志,满足生产环境监控需求。

🚀 快速开始

环境要求

  • Python 3.10+
  • Node.js 18+ 和 Bun
  • uv(推荐的 Python 包管理工具)

从源码安装

git clone https://github.com/RangerCao/LightRAG.git
cd LightRAG

# 配置 Python 环境
uv sync --extra api
source .venv/bin/activate        # Linux/macOS
# .venv\Scripts\activate         # Windows

# 构建前端
cd lightrag_webui
bun install --frozen-lockfile
bun run build
cd ..

# 配置环境变量
cp env.example .env
# 编辑 .env,填入你的 LLM 和 Embedding API 密钥

# 启动服务
lightrag-server

Docker 部署

git clone https://github.com/RangerCao/LightRAG.git
cd LightRAG
cp env.example .env
# 编辑 .env 配置文件
docker compose -f docker-compose-full.yml up -d

完整 Docker 技术栈包括:Mindage 服务、vLLM 嵌入/重排服务、PostgreSQL (pgvector)、Neo4j 图数据库和 Milvus 向量数据库。

访问 Mindage

启动后,在浏览器中打开:

  • WebUI:http://localhost:9621/webui/
  • API 文档:http://localhost:9621/docs
  • 健康检查:http://localhost:9621/health

⚙️ 配置说明

Mindage 通过 .env 文件中的环境变量进行配置。主要设置项:

变量说明默认值
LLM_BINDINGLLM 提供商(openai / ollama / azure)openai
LLM_MODELLLM 模型名称gpt-4o-mini
EMBEDDING_BINDING嵌入模型提供商openai
EMBEDDING_MODEL嵌入模型名称text-embedding-3-small
SUMMARY_LANGUAGE实体/关系输出语言English
ENABLE_PAGEINDEX启用树形索引功能false
MAX_PARALLEL_INSERT最大并行处理文档数3

完整配置请参考 env.example 文件。

🏗️ 项目架构

Mindage
├── lightrag/              # Python 核心包
│   ├── api/               # FastAPI 服务 + REST 接口
│   │   ├── routers/       # 文档、查询、图谱、探索、出行、MCP 路由
│   │   └── lightrag_server.py
│   ├── kg/                # 存储后端(Neo4j、PostgreSQL、Milvus 等)
│   ├── llm/               # LLM 提供商(OpenAI、Ollama、Azure 等)
│   ├── parser/            # 文档解析器(native、MinerU、Docling)
│   ├── chunker/           # 分块策略
│   ├── pipeline.py        # 文档摄入流水线
│   └── operate.py         # 抽取与检索操作
├── lightrag_webui/        # React 19 + TypeScript 前端
│   └── src/
│       ├── components/    # UI 组件
│       ├── views/         # 页面视图(对话、知识、图谱、思维导图、出行...)
│       └── i18n/          # 11 种语言翻译
└── docker-compose-full.yml

🙏 致谢

Mindage 基于 HKUDS 的 LightRAG 项目构建。我们诚挚感谢原项目在基于图谱的检索增强生成领域的基础性贡献。

如果您在研究中使用了 Mindage,请引用 LightRAG 原始论文:

@article{guo2024lightrag,
  title={LightRAG: Simple and Fast Retrieval-Augmented Generation},
  author={Zirui Guo and Lianghao Xia and Yanhua Yu and Tu Ao and Chao Huang},
  year={2024},
  eprint={2410.05779},
  archivePrefix={arXiv},
  primaryClass={cs.IR}
}

↑↓ navigate openesc close