mcp-0ne
A standalone MCP gateway that multiplexes multiple backend MCP servers behind a single endpoint, supporting both HTTP and stdio transports. It allows users to dynamically manage backends at runtime and organizes tools using a configurable namespace prefix.
README
mcp-0ne
One MCP server to rule them all.
A standalone MCP Gateway that multiplexes N backend MCP servers behind a single endpoint. Connect your MCP client once, manage backends dynamically at runtime.
MCP Client ──(HTTP)──> mcp-0ne (:8150)
│
┌───────────────┼───────────────┐
│ │ │
(HTTP client) (stdio spawn) (stdio spawn)
│ │ │
HTTP MCP stdio MCP stdio MCP
server server server
Quickstart
# Install
pip install -e .
# Start with empty backends
python -m mcp_0ne.server
# Verify
curl http://127.0.0.1:8150/health
Connect Claude Code
Add to your .mcp.json:
{
"mcpServers": {
"0ne": {
"type": "http",
"url": "http://127.0.0.1:8150/mcp"
}
}
}
Add Backends
Via MCP tool (from Claude Code)
Call 0ne__add:
{
"id": "myapp",
"type": "http",
"prefix": "myapp",
"url": "http://127.0.0.1:8100/mcp",
"description": "My App MCP Server"
}
Via REST API
curl -X POST http://127.0.0.1:8150/api/backends/myapp \
-H "Content-Type: application/json" \
-d '{"type":"http","prefix":"myapp","url":"http://127.0.0.1:8100/mcp"}'
Via config file
Edit backends.json:
{
"backends": {
"myapp": {
"type": "http",
"url": "http://127.0.0.1:8100/mcp",
"prefix": "myapp",
"enabled": true,
"description": "My App MCP Server"
}
},
"settings": {
"separator": "__",
"lazy_connect": true,
"tool_cache_ttl": 60,
"log_level": "info"
}
}
Backend Types
HTTP
Connects to an existing MCP server over HTTP JSON-RPC 2.0.
| Field | Required | Description |
|---|---|---|
type |
yes | "http" |
prefix |
yes | Tool namespace prefix |
url |
yes | MCP endpoint URL |
health_url |
no | Health check URL |
timeout |
no | Request timeout (default: 30s) |
description |
no | Human-readable description |
enabled |
no | Auto-connect (default: true) |
Stdio
Spawns a subprocess and communicates via MCP SDK stdio transport.
| Field | Required | Description |
|---|---|---|
type |
yes | "stdio" |
prefix |
yes | Tool namespace prefix |
command |
yes | Executable path |
args |
no | Command arguments |
env |
no | Environment variables |
timeout |
no | Operation timeout (default: 60s) |
description |
no | Human-readable description |
enabled |
no | Auto-connect (default: true) |
Admin Tools
Available as MCP tools (prefix 0ne__):
| Tool | Description |
|---|---|
0ne__discover |
List all backends with state and tool counts |
0ne__health |
Health check all backends with latency |
0ne__add |
Register a new backend at runtime |
0ne__remove |
Disconnect and unregister a backend |
0ne__enable |
Enable a disabled backend |
0ne__disable |
Disable a backend without removing it |
0ne__refresh |
Force reconnect and re-enumerate tools |
REST API
| Method | Path | Description |
|---|---|---|
GET |
/api/backends |
List all backends |
POST |
/api/backends/{id} |
Add a new backend |
DELETE |
/api/backends/{id} |
Remove a backend |
PATCH |
/api/backends/{id} |
Update backend config |
POST |
/api/backends/{id}/refresh |
Reconnect and refresh |
Environment Variables
| Variable | Default | Description |
|---|---|---|
MCP_0NE_HOST |
127.0.0.1 |
Server bind address |
MCP_0NE_PORT |
8150 |
Server port |
MCP_0NE_CONFIG |
./backends.json |
Config file path |
MCP_0NE_LOG_LEVEL |
info |
Log level |
Tool Namespacing
Tools from backends are exposed with a namespace prefix:
{prefix}__{original_tool_name}
For example, a backend with prefix myapp exposing a tool search becomes myapp__search. The separator __ is configurable in settings.
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。