SAG MCP Server
SAG MCP Server
Servidor MCP de lectura/escritura completo para la base de conocimiento de SAG. Expone 18 herramientas a través del transporte Streamable HTTP, compatible con Claude Desktop, Cursor, VS Code y cualquier cliente MCP.
Características
18 herramientas: CRUD de fuentes/documentos, búsqueda semántica, consultas al grafo de conocimiento, configuración de modelos
Streamable HTTP: conéctate desde cualquier cliente MCP sin dependencias locales
Autenticación Bearer: autenticación con token JWT integrada
Cero configuración: funciona con una instancia de SAG en ejecución sin ajustes adicionales
Related MCP server: LightRAG MCP Server
Inicio rápido
1. Requisitos previos
Una instancia de SAG en ejecución (Docker o nativa). Consulta la guía de despliegue de SAG.
2. Instalación
pip install sag-mcp-serverO desde el código fuente:
git clone https://github.com/Zleap-AI/sag-mcp-server.git
cd sag-mcp-server
pip install -e .3. Ejecutar
Modo HTTP (recomendado):
export SAG_API_URL=http://your-sag-host:9002
sag-mcp-server http 9003Modo stdio (para integración local):
sag-mcp-server stdio4. Conectar desde Claude Desktop
Añade a claude_desktop_config.json:
{
"mcpServers": {
"sag": {
"type": "http",
"url": "http://your-sag-host:9003/mcp/",
"headers": {
"Authorization": "Bearer <your-jwt-token>"
}
}
}
}Obtén un token:
curl -s -X POST http://your-sag-host:9002/api/v1/auth/login \
-H 'Content-Type: application/json' -d '{"name":"Admin"}' \
| python3 -c "import sys,json;print(json.load(sys.stdin)['access_token'])"Herramientas
Categoría | Herramienta | Descripción |
Fuentes |
| Lista todas las fuentes de la base de conocimiento |
| Crea una nueva fuente | |
| Elimina una fuente y todos sus datos | |
Documentos |
| Lista los documentos de una fuente |
| Sube un archivo local (autoanaliza y extrae) | |
| Escribe texto directamente (sin necesidad de archivo) | |
| Reprocesa un documento | |
| Pausa el procesamiento | |
| Reanuda el procesamiento | |
| Elimina un documento | |
Recuperación |
| Búsqueda semántica con resumen LLM |
| Consulta una entidad del grafo de conocimiento | |
| Obtiene el esquema del documento | |
| Búsqueda por palabra clave o expresión regular | |
| Lee el texto completo del documento | |
Configuración |
| Ver la configuración del modelo |
| Actualizar los ajustes del modelo | |
| Probar la conectividad del modelo |
Configuración
Toda la configuración se realiza mediante variables de entorno:
Variable | Por defecto | Descripción |
|
| URL base de la API de SAG |
| (vacío = auto-login) | Token Bearer JWT |
|
| Nombre de usuario (cuando el token está vacío) |
|
| Puerto del servidor HTTP |
Configuración del cliente
Claude Desktop
{
"mcpServers": {
"sag": {
"type": "http",
"url": "http://your-sag-host:9003/mcp/",
"headers": { "Authorization": "Bearer <token>" }
}
}
}Cursor
~/.cursor/mcp.json:
{
"mcpServers": {
"sag": {
"type": "http",
"url": "http://your-sag-host:9003/mcp/",
"headers": { "Authorization": "Bearer <token>" }
}
}
}Dify
Tipo: Streamable HTTP
URL:
http://your-sag-host:9003/mcp/Cabecera:
Authorization: Bearer <token>
Docker
Ejecuta como contenedor junto a SAG:
# docker-compose.yml addition
services:
mcp:
build: .
environment:
SAG_API_URL: http://api:8000
SAG_SECRET_KEY: ${SAG_SECRET_KEY}
ports:
- "9003:9003"
command: ["sag-mcp-server", "http", "9003"]
depends_on:
api:
condition: service_healthyDesarrollo
git clone https://github.com/Zleap-AI/sag-mcp-server.git
cd sag-mcp-server
pip install -e ".[dev]"
# Run tests
pytest
# Lint
ruff check src/Compatibilidad de API
Requiere SAG v0.7.x+ con los siguientes endpoints:
POST /api/v1/auth/login— Autenticación JWTGET/POST/DELETE /api/v1/sources— CRUD de fuentesGET/POST/DELETE /api/v1/sources/{id}/documents— CRUD de documentosPOST /api/v1/search— Búsqueda semánticaGET /api/v1/system/model-config— Configuración del modelo
Licencia
Licencia MIT. Consulta LICENSE para más detalles.
中文文档
基于 SAG 知识库的读写 MCP Server。通过 Streamable HTTP 暴露 18 个工具,兼容 Claude Desktop、Cursor、VS Code 及所有 MCP 客户端。
特性
18 个工具:信源/文档增删改查、语义检索、知识图谱查询、模型配置
Streamable HTTP:无需本地安装任何依赖,客户端通过 HTTP 直连
Bearer 鉴权:内置 JWT token 认证
零配置:与运行中的 SAG 实例直接配合使用
快速开始
1. 前提条件
运行中的 SAG 实例(Docker 或原生部署)。参见 SAG 部署指南。
2. 安装
pip install sag-mcp-server或从源码安装:
git clone https://github.com/Zleap-AI/sag-mcp-server.git
cd sag-mcp-server
pip install -e .3. 启动
HTTP 模式(推荐):
export SAG_API_URL=http://your-sag-host:9002
sag-mcp-server http 9003stdio 模式(本地集成):
sag-mcp-server stdio4. 连接 Claude Desktop
编辑 claude_desktop_config.json:
{
"mcpServers": {
"sag": {
"type": "http",
"url": "http://your-sag-host:9003/mcp/",
"headers": {
"Authorization": "Bearer <your-jwt-token>"
}
}
}
}获取 token:
curl -s -X POST http://your-sag-host:9002/api/v1/auth/login \
-H 'Content-Type: application/json' -d '{"name":"Admin"}' \
| python3 -c "import sys,json;print(json.load(sys.stdin)['access_token'])"工具清单
类别 | 工具 | 说明 |
信源 |
| 列出所有知识库信源 |
| 创建新信源 | |
| 删除信源及全部数据 | |
文档 |
| 列出信源下文档 |
| 上传本地文件(自动解析+抽取) | |
| 直接写入文本(无需文件) | |
| 重新处理文档 | |
| 暂停处理 | |
| 恢复处理 | |
| 删除文档 | |
检索 |
| 语义检索(含 LLM 摘要) |
| 查询知识图谱实体 | |
| 获取文档大纲 | |
| 关键词/正则检索 | |
| 读取文档全文 | |
模型 |
| 查看模型配置 |
| 更新模型设置 | |
| 测试模型连通性 |
配置说明
所有配置通过环境变量:
变量 | 默认值 | 说明 |
|
| SAG API 地址 |
| (空=自动登录) | JWT Bearer token |
|
| 登录用户名(token 为空时生效) |
|
| HTTP 服务端口 |
客户端配置
Claude Desktop
{
"mcpServers": {
"sag": {
"type": "http",
"url": "http://your-sag-host:9003/mcp/",
"headers": { "Authorization": "Bearer <token>" }
}
}
}Cursor
~/.cursor/mcp.json:
{
"mcpServers": {
"sag": {
"type": "http",
"url": "http://your-sag-host:9003/mcp/",
"headers": { "Authorization": "Bearer <token>" }
}
}
}Dify
类型:Streamable HTTP
URL:
http://your-sag-host:9003/mcp/请求头:
Authorization: Bearer <token>
Docker 部署
与 SAG 一起以容器方式运行:
# 添加到 docker-compose.yml
services:
mcp:
build: .
environment:
SAG_API_URL: http://api:8000
SAG_SECRET_KEY: ${SAG_SECRET_KEY}
ports:
- "9003:9003"
command: ["sag-mcp-server", "http", "9003"]
depends_on:
api:
condition: service_healthy开发指南
git clone https://github.com/Zleap-AI/sag-mcp-server.git
cd sag-mcp-server
pip install -e ".[dev]"
# 运行测试
pytest
# 代码检查
ruff check src/API 兼容性
需要 SAG v0.7.x+,支持以下接口:
POST /api/v1/auth/login— JWT 认证GET/POST/DELETE /api/v1/sources— 信源增删改查GET/POST/DELETE /api/v1/sources/{id}/documents— 文档增删改查POST /api/v1/search— 语义检索GET /api/v1/system/model-config— 模型配置
许可证
MIT License。详见 LICENSE。
This server cannot be installed
Maintenance
Resources
Unclaimed servers have limited discoverability.
Looking for Admin?
If you are the server author, to access and configure the admin panel.
Related MCP Servers
- AlicenseNot gradedqualityFmaintenanceA comprehensive MCP server that provides full integration with LightRAG API, offering 22 tools across document management, querying, knowledge graph operations, and system management.37MIT
- AlicenseNot gradedqualityBmaintenanceEnables AI clients to interact with a LightRAG knowledge graph server via MCP, providing 30 tools for queries, document management, and graph operations.283MIT
- FlicenseNot gradedqualityCmaintenanceEnables building and querying knowledge graphs by ingesting documents into Neo4j using Gemini for entity extraction, and exposes MCP tools for graph health, document ingestion, and knowledge base querying.
- AlicenseNot gradedqualityBmaintenanceEnables document-based Q&A with multi-modal RAG, hybrid retrieval, knowledge graph reasoning, and multi-agent orchestration via MCP tools.4MIT
Related MCP Connectors
Knowledge coverage map and health score. Ingest docs into a governed knowledge graph via MCP.
Manage SRG+ hubs, channels, content, assets, users, and workspaces from any MCP-aware AI agent.
Shared, governed long-term memory for AI agents across tools and sessions via MCP and REST.
Latest Blog Posts
- Who's Calling? MCP Hosts Are an Identity Blind Spot (And the Spec Knows It)By Om-Shree-0709 on .mcpAgent IdentityOAuth 2.1
- Your AI Chatbot Just Exposed Your CEO's Salary to an InternBy Om-Shree-0709 on .Agent IdentityMCP SecurityOAuth Delegation
- Why MCP Servers Need Execution Sandboxing (And Why Your Current Stack Isn't Enough)By Om-Shree-0709 on .Agentic AiPrompt InjectionWebAssembly
MCP directory API
We provide all the information about MCP servers via our MCP API.
curl -X GET 'https://glama.ai/api/mcp/v1/servers/guazixiong/sag-mcp-server'
If you have feedback or need assistance with the MCP directory API, please join our Discord server