McpMux

平台与服务

by mcpmux

McpMux 让你一次配置 MCP servers,即可通过一个本地 gateway 连接所有 AI client,显著简化接入。

什么是 McpMux

McpMux 让你一次配置 MCP servers,即可通过一个本地 gateway 连接所有 AI client,显著简化接入。

README

McpMux

License: GPL v3 GitHub release CI Windows macOS Linux

One app to manage all your MCP servers across every AI client.

Website · Download · Discover Servers · Features · Discord


The Problem

Cursor, Claude Desktop, VS Code, Windsurf — they all support MCP, but each one needs its own config file. None of them talk to each other.

code
┌─────────────────────────────────────────────────────────────────────────┐
│  Today: every client manages MCP servers independently                  │
│                                                                         │
│  Cursor       → config.json    → github, slack, db  + API keys         │
│  Claude       → config.json    → github, slack, db  + API keys  (dupe) │
│  VS Code      → settings.json  → github, slack, db  + API keys  (dupe) │
│  Windsurf     → config.json    → github, slack, db  + API keys  (dupe) │
│                                                                         │
│  ⚠ 4 config files  ·  4 copies of every API key  ·  all plain text     │
└─────────────────────────────────────────────────────────────────────────┘

Add a server? Edit four files. Rotate an API key? Edit four files. New machine? Start from scratch.

And those API keys? Sitting in plain-text JSON files anyone on your machine can read.

The Fix

McpMux is a desktop app that runs a local MCP gateway. You configure servers once — every AI client connects through a single URL.

code
┌─────────────────────────────────────────────────────────────────────────┐
│  With McpMux: one gateway, every client                                 │
│                                                                         │
│  Cursor       ─┐                                                        │
│  Claude       ─┤                  ┌──→  GitHub                          │
│  VS Code      ─┼──→  McpMux      ├──→  Slack                           │
│  Windsurf     ─┤   localhost:45818├──→  PostgreSQL                      │
│  Any client   ─┘                  └──→  + all your servers              │
│                                                                         │
│  ✓ 1 config  ·  credentials encrypted in OS keychain  ·  instant sync  │
└─────────────────────────────────────────────────────────────────────────┘

Add a server in McpMux and every connected client picks it up instantly — no restart, no manual refresh. Remove a tool or update a prompt? Every client knows immediately.

Lightweight and cross-platform — built in Rust with Tauri 2, McpMux uses minimal CPU and memory while running quietly in your system tray on Windows, macOS, and Linux.

McpMux Dashboard


How It Works

1. Install servers from the built-in registry — or add your own

2. Paste one config into each AI client (the last config you'll ever need):

json
{
  "mcpServers": {
    "mcpmux": {
      "type": "http",
      "url": "http://localhost:45818/mcp"
    }
  }
}

3. Done. Every tool from every server is available in every client, right now.

McpMux routes calls to the right server, refreshes OAuth tokens automatically, and keeps credentials encrypted in your OS keychain — you never think about it again. It also keeps itself current: new versions download and install on launch by default (toggle in Settings), so a restart is all it takes.


Features

All Your Servers, One Place

No more duplicating server configs across Cursor, Claude, VS Code, and Windsurf. Install a server in McpMux and it's immediately available everywhere. Expand any server to inspect its tools, prompts, and resources. See live connection status. OAuth tokens refresh automatically in the background.

Server Management — Azure, AWS, Notion, Docker, Slack all connected

100+ Servers, One Click

Stop hunting for MCP server repos and hand-writing transport configs. Browse a curated registry of 100+ servers — GitHub, Slack, PostgreSQL, Docker, Notion, AWS, Azure, and more. Click install, enter your credentials, and the server is live across every AI client you use. You can also browse the full registry at mcpmux.com.

Discover — browse 100+ servers and install with one click

The full registry is also available on the web at mcpmux.com — with search, categories, and one-click install via deep links.

mcpmux.com — server registry on the web

Workspaces That Keep Things Separate

Create isolated Spaces — each with their own servers, credentials, and permissions. A "Work" space for company databases and internal APIs. A "Personal" space for side projects. Switch in one click from the sidebar and every connected AI client follows automatically. No more accidentally querying your personal database from a work project.

Workspaces — switch context instantly from the sidebar

Different Tools for Different Folders

Your AI client tells McpMux which folder it's working in (its MCP root). McpMux uses that to route each workspace to its own toolset — open your backend repo and the AI sees your database and deploy tools; open a docs folder and it sees only search and filesystem. Map a folder once in the Workspaces tab (or let the AI do it — see below) and every future session from that exact path resolves automatically. Matching is per-folder and exact, so nothing leaks across projects.

Workspaces — map a folder to the Space and FeatureSet it should get

Control What Each Client Can Do

Not every AI client should have the same power. Create Feature Sets — curated bundles that control exactly which tools, prompts, and resources are exposed. Build a "Read Only" set for cautious workflows, a "React Development" set with just GitHub and Filesystem, or a "Full Stack Dev" set with everything — then route a folder to it via a Workspace mapping. A FeatureSet's included features are the effective toolset a session resolves to.

Feature Sets — pick exactly which tools each bundle exposes, per server

See and Manage Every Connected Client

Cursor, VS Code, Windsurf, Claude Code — see every AI app connected to your gateway in real time, with live status. Routing is workspace-driven: each app's toolset is decided by the Workspace binding for the folder it reports, not configured per app. Open any app to rename it, see how it's routed, or revoke its connection. New apps authenticate via OAuth with a one-click approval flow.

Connected apps — routing is workspace-driven per reported folder

Let Your AI Curate Its Own Toolset

Hand an assistant a hundred tools and it burns tokens and reaches for the wrong one. McpMux ships a built-in Tool Optimization capability so the AI can keep itself lean — straight from chat, no config files.

Start a request with @mux and the assistant can:

  • Discover what's available — mcpmux_list_spaces, mcpmux_list_all_tools, mcpmux_search_tools
  • Compose a focused FeatureSet of just the tools it needs — mcpmux_manage_feature_set
  • Pin the current folder to that set so it sticks — mcpmux_bind_current_workspace

Tool Optimization — the built-in self-management tools the AI drives, reads silent and writes gated

Reads are silent; anything that changes your setup pops a one-click approval dialog that names the exact Space — the AI proposes, you decide. The @mux trigger keeps these requests cleanly separated from your real work, and every operation can target a specific Space by id.

Approval — every self-management write asks first, showing the target Space and the exact tool diff

"@mux build a minimal toolset for this Next.js repo and pin it to this folder."


Security

MCP defaults to plain-text config files with raw API keys. McpMux replaces that with defense in depth:

  • OS Keychain — secrets in platform-native secure storage (DPAPI on Windows, Keychain on macOS, libsecret on Linux), never in plain-text files
  • AES-256-GCM — field-level database encryption for all sensitive data
  • OAuth 2.1 + PKCE — standard auth flow with automatic token refresh
  • Local-only gateway — binds to 127.0.0.1, nothing exposed to the network
  • Per-client access keys — each AI client authenticates independently with granular permissions
  • Sanitized logs — tokens and secrets never appear in log output
  • Memory zeroization — secrets wiped from memory after use via zeroize

All MCP traffic stays on your machine. Cloud sync (optional, coming soon) only covers config metadata — never credentials or payloads.


Getting Started

1. Download McpMux — Windows, macOS, Linux

2. Add servers — use the Discover tab to browse 100+ servers, or explore servers on mcpmux.com and install with one click

3. Paste the config — copy the snippet from the Dashboard into your AI clients

That's the last config file you'll ever touch.

Linux quick install: curl -fsSL https://install.mcpmux.com | bash

macOS via Homebrew: brew install --cask mcpmux/tap/mcpmux

See mcpmux.com/download for all platforms and install methods.


Install on Linux

Quick install (detects your distro automatically):

bash
curl -fsSL https://install.mcpmux.com | bash
<details> <summary>Other install methods</summary>

Debian / Ubuntu (APT repository for automatic updates):

bash
curl -fsSL https://install.mcpmux.com/apt | sudo bash

Fedora / RHEL (from GitHub Releases):

bash
# Download the latest .rpm
sudo dnf install https://github.com/mcpmux/mcp-mux/releases/latest/download/mcpmux-0.0.12-1.amd64.rpm

Arch Linux (AUR):

bash
yay -S mcpmux-bin

AppImage (any distro):

bash
curl -fsSL https://install.mcpmux.com | bash  # auto-detects, falls back to AppImage

Or download directly from GitHub Releases.

</details>

Development

bash
pnpm setup    # First-time: install dependencies
pnpm dev      # Start development
pnpm build    # Production build
pnpm test     # Run all tests

Prerequisites: Rust 1.75+, Node.js 18+, pnpm 9+. Linux also needs gnome-keyring libsecret-1-dev librsvg2-dev pkg-config.

Built with Tauri 2 (Rust + React 19), Axum for the gateway, ring for encryption, rmcp for MCP.

<details> <summary>Project structure</summary>
code
mcp-mux/
├── apps/desktop/          # Tauri desktop app (React + Rust)
├── crates/
│   ├── mcpmux-core/       # Domain logic
│   ├── mcpmux-gateway/    # HTTP gateway, OAuth, routing
│   ├── mcpmux-storage/    # SQLite + encryption + OS keychain
│   └── mcpmux-mcp/        # MCP protocol
├── packages/ui/           # Shared UI components
└── tests/                 # Unit, integration, E2E tests
</details>

Links

Contributing

See CONTRIBUTING.md for guidelines.

License

GNU General Public License v3.0

常见问题

McpMux 是什么?

McpMux 让你一次配置 MCP servers,即可通过一个本地 gateway 连接所有 AI client,显著简化接入。

相关 Skills

Slack动图

by anthropics

Universal
热门

面向Slack的动图制作Skill,内置emoji/消息GIF的尺寸、帧率和色彩约束、校验与优化流程,适合把创意或上传图片快速做成可直接发送的Slack动画。

帮你快速做出适配 Slack 的动图,内置约束规则和校验工具,少踩上传与播放坑,做表情包和演示都更省心。

平台与服务
未扫描164.6k

MCP构建

by anthropics

Universal
热门

聚焦高质量 MCP Server 开发,覆盖协议研究、工具设计、错误处理与传输选型,适合用 FastMCP 或 MCP SDK 对接外部 API、封装服务能力。

想让 LLM 稳定调用外部 API,就用 MCP构建:从 Python 到 Node 都有成熟指引,帮你更快做出高质量 MCP 服务器。

平台与服务
未扫描164.6k

接口测试套件

by alirezarezvani

Universal
热门

扫描 Next.js、Express、FastAPI、Django REST 的 API 路由,自动生成覆盖鉴权、参数校验、错误码、分页、上传与限流场景的 Vitest 或 Pytest 测试套件。

帮你把API与集成测试自动化跑顺,减少回归漏测;能力全面,尤其适合复杂接口场景的QA团队。

平台与服务
未扫描23.3k

相关 MCP Server

Slack 消息

编辑精选

by Anthropic

热门

Slack 是让 AI 助手直接读写你的 Slack 频道和消息的 MCP 服务器。

这个服务器解决了团队协作中需要 AI 实时获取 Slack 信息的痛点,特别适合开发团队让 Claude 帮忙汇总频道讨论或发送通知。不过,它目前只是参考实现,文档有限,不建议在生产环境直接使用——更适合开发者学习 MCP 如何集成第三方服务。

平台与服务
89.0k

by netdata

热门

io.github.netdata/mcp-server 是让 AI 助手实时监控服务器指标和日志的 MCP 服务器。

这个工具解决了运维人员需要手动检查系统状态的痛点,最适合 DevOps 团队让 Claude 自动分析性能数据。不过,它依赖 NetData 的现有部署,如果你没用过这个监控平台,得先花时间配置。

平台与服务
79.5k

by d4vinci

热门

Scrapling MCP Server 是专为现代网页设计的智能爬虫工具,支持绕过 Cloudflare 等反爬机制。

这个工具解决了爬取动态网页和反爬网站时的头疼问题,特别适合需要批量采集电商价格或新闻数据的开发者。不过,它依赖外部浏览器引擎,资源消耗较大,不适合轻量级任务。

平台与服务
71.5k

评论