SiYuan MCP Server
Enables AI assistants like Claude to interact with your SiYuan knowledge base through a secure, authenticated API.
README
SiYuan MCP Server
A Model Context Protocol (MCP) server for SiYuan Note with pluggable authentication. Enables AI assistants like Claude to interact with your SiYuan knowledge base through a secure, authenticated API.
Features
- Full SiYuan Integration: Read, write, search, and manage documents, blocks, flashcards, and more
- Pluggable Authentication:
- OAuth 2.1 + PKCE via Cloudflare Access (supports Okta, Azure AD, Google, etc.)
- Simple API key authentication via
X-SiYuan-Keyheader
- Multi-Worker Architecture: Separate auth and MCP backend for flexibility
- Two Deployment Modes:
- Cloudflare Workers: Production deployment with multiple auth options
- CLI (stdio): Standalone MCP server for direct Claude Desktop integration
- RAG Support: Optional vector search integration for semantic document retrieval
- Read-Only Mode: Configurable restrictions for safe read-only access
Architecture
┌─────────────────────────────────────────────────────────────────────────┐
│ Cloudflare Workers Mode │
├─────────────────────────────────────────────────────────────────────────┤
│ │
│ ┌─────────────────────────────┐ ┌─────────────────────────────┐ │
│ │ CF Access Auth (sy.wenri.org)│ │ API Key Auth (api-sy.wenri.org)│
│ │ - OAuth flow (/authorize) │ │ - X-SiYuan-Key validation │ │
│ │ - /download (grant-based) │ │ - /download (stateless) │ │
│ └──────────────┬──────────────┘ └──────────────┬──────────────┘ │
│ │ Service Binding │ Service Binding │
│ └───────────────────┬───────────────┘ │
│ ▼ │
│ ┌───────────────────────────────┐ │
│ │ MCP Backend Worker │ │
│ │ - SiyuanMCP Durable Object │ │
│ │ - Tool execution │ │
│ │ - SiYuan Kernel API calls │ │
│ └───────────────────────────────┘ │
└─────────────────────────────────────────────────────────────────────────┘
┌─────────────────────────────────────────────────────────────────────────┐
│ CLI Mode (stdio) │
├─────────────────────────────────────────────────────────────────────────┤
│ Claude Desktop ←──stdio──→ handlers/cli.ts ←──HTTP──→ SiYuan Kernel │
└─────────────────────────────────────────────────────────────────────────┘
Available MCP Tools
| Tool Category | Tools |
|---|---|
| Document Read | List notebooks, get document tree, read document content, outline |
| Document Write | Create, rename, move, delete documents |
| Block Operations | Insert, update, delete, move blocks (with batch support) |
| Search | Full-text search (siyuan_find_block), SQL queries |
| Vector Search | RAG-based semantic search (requires RAG backend) |
| Daily Notes | Create and manage daily notes |
| Flashcards | Create and review flashcards |
| Attributes | Manage custom attributes on documents/blocks |
| Relations | Manage block relations |
| Assets | Upload assets (batch, URL fetch, JSON auto-serialize) |
| File System | Read/write files, create archives |
| Templates | Render and manage SiYuan templates |
| Help Docs | Built-in documentation resources |
| Utilities | Get time, push notifications, reindex, flush transactions |
Quick Start
Option 1: CLI Mode (Local Development)
Use stdio transport for direct Claude Desktop integration:
# Clone and install
git clone <repo-url>
cd mcp_saas
npm install
# Run with SiYuan kernel URL
npx tsx handlers/cli.ts --kernel-url http://localhost:6806
# Or with authentication token
npx tsx handlers/cli.ts --kernel-url http://localhost:6806 --token YOUR_TOKEN
Add to Claude Desktop (claude_desktop_config.json):
{
"mcpServers": {
"siyuan": {
"command": "npx",
"args": ["tsx", "/path/to/handlers/cli.ts", "--kernel-url", "http://localhost:6806"]
}
}
}
Option 2: Cloudflare Workers (Production)
Deploy multi-worker MCP server to Cloudflare Workers.
Project Structure
workers/
├── mcp-backend/ # MCP Backend (internal, no public routes)
│ ├── index.ts # Entry point, WorkerEntrypoint with RPC methods
│ ├── server/ # MCP server core
│ │ ├── agent.ts # SiyuanMCP Durable Object
│ │ └── index.ts # Server initialization
│ └── wrangler.jsonc # DO bindings
│
├── auth-cfaccess/ # CF Access OAuth (sy.wenri.org)
│ ├── index.ts # OAuthProvider + RPC forwarding
│ ├── access-handler.tsx # Hono app: OAuth flow, consent (JSX)
│ └── wrangler.jsonc # KV + service binding
│
└── auth-apikey/ # API Key Auth (api-sy.wenri.org)
├── index.ts # Hono app: X-SiYuan-Key validation
└── wrangler.jsonc # Service binding
Deploy MCP Backend (First)
cd workers/mcp-backend
wrangler secret put SIYUAN_KERNEL_TOKEN
# If SiYuan kernel is behind CF Access:
wrangler secret put CF_ACCESS_SERVICE_CLIENT_ID
wrangler secret put CF_ACCESS_SERVICE_CLIENT_SECRET
npx wrangler deploy
Deploy CF Access Auth Worker
cd workers/auth-cfaccess
# Create KV namespace (one-time)
npx wrangler kv namespace create "OAUTH_KV"
# Update KV ID in wrangler.jsonc
# Set secrets from CF Access SaaS app dashboard
wrangler secret put ACCESS_CLIENT_ID
wrangler secret put ACCESS_CLIENT_SECRET
wrangler secret put ACCESS_TOKEN_URL
wrangler secret put ACCESS_AUTHORIZATION_URL
wrangler secret put ACCESS_JWKS_URL
wrangler secret put COOKIE_ENCRYPTION_KEY # openssl rand -hex 32
npx wrangler deploy
Deploy API Key Auth Worker
cd workers/auth-apikey
wrangler secret put SIYUAN_KERNEL_TOKEN
wrangler secret put COOKIE_ENCRYPTION_KEY
npx wrangler deploy
Connect with Claude Desktop
Via OAuth (CF Access):
{
"mcpServers": {
"siyuan-oauth": {
"command": "npx",
"args": ["mcp-remote", "https://sy.wenri.org/sse"]
}
}
}
Via API Key:
claude mcp add siyuan https://api-sy.wenri.org/sse \
-t sse -H "X-SiYuan-Key: YOUR_TOKEN"
Configuration Reference
Environment Variables
| Variable | Required | Description |
|---|---|---|
SIYUAN_KERNEL_URL |
Yes | SiYuan kernel URL |
SIYUAN_KERNEL_TOKEN |
If auth enabled | SiYuan API token |
COOKIE_ENCRYPTION_KEY |
Workers mode | For download URL encryption |
RAG_BASE_URL |
Optional | RAG backend URL for vector search |
RAG_API_KEY |
Optional | RAG backend API key |
FILTER_NOTEBOOKS |
Optional | Newline-separated notebook IDs to include |
FILTER_DOCUMENTS |
Optional | Newline-separated document IDs to include |
READ_ONLY_MODE |
Optional | allow_all, allow_non_destructive, or deny_all |
AUTO_APPROVE_LOCAL_CHANGE |
Optional | Auto-approve local change operations |
CF Access Auth Worker Secrets
| Secret | Description |
|---|---|
ACCESS_CLIENT_ID |
From CF Access SaaS app dashboard |
ACCESS_CLIENT_SECRET |
From CF Access SaaS app dashboard |
ACCESS_TOKEN_URL |
Token endpoint URL |
ACCESS_AUTHORIZATION_URL |
Authorization endpoint URL |
ACCESS_JWKS_URL |
JWKS endpoint URL |
CLI Options
Options:
-u, --kernel-url <url> SiYuan kernel URL (required)
-t, --token <token> SiYuan API token
--rag-url <url> RAG backend URL
--rag-key <key> RAG API key
--filter-notebooks <ids> Notebook IDs to filter (newline-separated)
--filter-documents <ids> Document IDs to filter (newline-separated)
--read-only <mode> Read-only mode: allow_all, allow_non_destructive, deny_all
-h, --help Show help message
API Endpoints
OAuth Auth Worker (sy.wenri.org)
GET /authorize- Initiate OAuth flowGET /callback- OAuth callback, consent pagePOST /callback- Complete authorizationPOST /token- Token endpointPOST /register- Dynamic client registrationGET /.well-known/oauth-authorization-server- OAuth metadataPOST /mcp,GET /sse- MCP endpoints (forwarded to backend via RPC)GET /download/*- File downloads (grant-based validation)
API Key Auth Worker (api-sy.wenri.org)
POST /mcp,GET /sse- MCP endpoints (X-SiYuan-Key required, forwarded via RPC)GET /download/*- File downloads (stateless validation)
Development
# Install dependencies
npm install
# Local development - start each worker separately
cd workers/mcp-backend && npx wrangler dev # http://localhost:8787
cd workers/auth-cfaccess && npx wrangler dev # http://localhost:8788
cd workers/auth-apikey && npx wrangler dev # http://localhost:8789
# Run tests
npm test
# Deploy (order matters: backend first)
cd workers/mcp-backend && npx wrangler deploy
cd workers/auth-cfaccess && npx wrangler deploy
cd workers/auth-apikey && npx wrangler deploy
Testing
MCP Inspector
npx @modelcontextprotocol/inspector@latest
# Connect to deployed URL or localhost
Manual Testing
# Test OAuth discovery
curl https://sy.wenri.org/.well-known/oauth-authorization-server
# Test API key auth
curl -X POST https://api-sy.wenri.org/mcp \
-H "X-SiYuan-Key: YOUR_TOKEN" \
-H "Content-Type: application/json" \
-d '{"jsonrpc":"2.0","method":"tools/list","id":1}'
Security
- OAuth 2.1 + PKCE: Prevents authorization code interception
- Cloudflare Access: Enterprise identity provider support
- Service Bindings: Internal worker communication (no public routes for backend)
- Durable Objects: Session state with SQLite storage
- Download URL Encryption: Time-bound, path-bound download tokens
- Read-Only Mode: Optional restriction of write operations
Troubleshooting
Common Issues
"SIYUAN_KERNEL_URL not configured"
- Set
SIYUAN_KERNEL_URLin wrangler.jsonc vars
"Failed to get SiYuan config"
- Verify SiYuan kernel is running and accessible
- Check
SIYUAN_KERNEL_TOKENif authentication is enabled
"Unauthorized: Missing auth context"
- MCP backend requires auth headers from auth workers
- Cannot be accessed directly; use auth worker endpoints
"Invalid or expired state"
- OAuth state expired (10 min timeout)
- Verify KV namespace is configured correctly
Tool not appearing
- Check
READ_ONLY_MODEsetting - Verify tool annotations allow current mode
License
MIT
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。