English | 中文

One local proxy that connects MCP servers and AI clients.

Configuring the same MCP servers across multiple clients is repetitive, token-costly, and hard to observe. MCPMate proxies MCP servers, syncs client configs, trims capabilities by profile, and logs activity.

This is not a brand-new project. I started shaping MCPMate around May 2024, paused active development around October, and recently came back to it with a clearer conviction: as the hype around skills- and CLI-shaped workflows settles into a more reflective phase, the long-term, irreplaceable value of MCP becomes easier to see.

MCPMate was previously developed in private and is now being reopened in public. The direction I care about most at this stage is usability: building on MCPMate's earlier profile-based approach for removing redundant capabilities in specific scenarios, and continuing to extend its hosted mode toward a more progressively disclosed Unify mode (last year I referred to it as a more "aggressive hosted" mode, though the name itself felt somewhat awkward). One goal is to bring some of the lower-friction and lower first-token-cost qualities that people appreciated in skills- and CLI-shaped experiences into MCP itself.

• MCPMate
  • 📑 Table of Contents
  • 🤔 Why MCPMate?
  • 🔄 How It Works
  • 🚀 Key Features
  • 🛠️ Core Components
    • Proxy
    • Bridge
    • Runtime Manager
    • Desktop App
    • Logs
  • ⚡ Quick Start
    • Option A: Download the Desktop App (Recommended)
    • Option B: Build from Source
    • Option C: Online Demo
  • 🧰 Tech Stack
  • 🚢 Deployment Modes
  • 🔧 Development
  • 🗺️ Roadmap
  • 🤝 Contributing
  • 📄 License

Managing MCP servers across multiple AI tools (Claude Desktop, Cursor, Zed, Codex, and user-defined clients) brings significant challenges:

flowchart LR
    subgraph mgmt ["🖥️ Management"]
        APP["Tauri Desktop"]
        DASH["Web Dashboard"]
    end

    subgraph proxy ["⚙️ MCPMate Proxy"]
        direction TB
        UP["Upstream Mgmt<br/><small>Connection Pool · OAuth 2.0 + PKCE</small>"]
        PROF["Profile Engine<br/><small>Scene · App · Dynamic</small>"]
        SEC["Event Logs &amp; Security<br/><small>Structured Logs · redb Cache</small>"]
        UP --> PROF --> SEC
    end

    subgraph cfg ["🔧 Client Config"]
        direction TB
        DET["Detection &amp; Adapter"]
        MODES["Modes<br/><small>Transparent · Hosted · Unify</small>"]
        DET --> MODES
    end

    subgraph clients ["📱 AI Clients"]
        C["Cursor · VS Code · Zed · ..."]
    end

    subgraph upstream ["☁️ Upstream MCP Servers"]
        S["uvx · bunx · npx<br/>SSE · Streamable HTTP"]
    end

    subgraph market ["🛒 MCP Marketplace"]
        M["Official Registry"]
    end

    APP -->|"API"| proxy
    DASH -->|"API"| proxy
    market -.->|"Install"| proxy
    upstream -->|"stdio / HTTP"| proxy
    proxy -->|"Bridge / HTTP"| cfg
    MODES --> clients

    classDef mgmt fill:#e3f2fd,stroke:#1565c0,color:#000
    classDef proxy fill:#fff3e0,stroke:#e65100,color:#000
    classDef cfg fill:#e8f5e9,stroke:#2e7d32,color:#000
    classDef clients fill:#fce4ec,stroke:#c62828,color:#000
    classDef upstream fill:#f3e5f5,stroke:#6a1b9a,color:#000
    classDef market fill:#fff9c4,stroke:#f9a825,color:#000

    class APP,DASH mgmt
    class UP,PROF,SEC proxy
    class DET,MODES cfg
    class C clients
    class S upstream
    class M market

MCPMate sits between your AI clients and MCP servers as a transparent proxy. The Bridge adapts stdio-only clients (like Claude Desktop) to the HTTP proxy. The Profile Engine controls which tools are visible to which client — scene profiles for workflow context, app profiles for per-client tuning, and dynamic profiles that adjust at runtime. The client configuration layer covers Transparent, Hosted, and Unify management modes.

A high-performance MCP proxy server that connects to multiple MCP servers and aggregates their tools. Implements stdio and Streamable HTTP transport protocols (aligned with MCP 2025-06-18 specification). Accepts legacy SSE-configured servers and automatically normalizes them to Streamable HTTP for backward compatibility.

A lightweight bridging component that converts stdio protocol to HTTP transport without modifying the client. Automatically inherits all functions and tools from the HTTP service. Minimal configuration — only requires service address.

Installs and manages runtimes used by local MCP servers. Supports Node.js, uv (Python), and Bun with download progress tracking and automatic environment variable configuration.

runtime install node   # Install Node.js for JavaScript MCP servers
runtime install uv     # Install uv for Python MCP servers
runtime install bun    # Install Bun
runtime list           # List installed runtimes

Cross-platform desktop application built with Tauri 2. Complete graphical interface for managing MCP servers, profiles, and tools with real-time monitoring, intelligent client detection, and system tray integration. macOS is available now; Windows is in beta; Linux is in development.

Structured operational log for MCP proxy activity. Collects MCP operations and management-side changes into a structured timeline with cursor-based pagination, REST APIs, and a dedicated Logs page in the dashboard UI.

Download the latest release for your platform from GitHub Releases.

Note: macOS builds are currently unsigned and not notarized. You may need to right-click and select "Open" to bypass Gatekeeper on first launch. Code signing and notarization are planned for a future release.

Prerequisites: Rust 1.75+, Node.js 18+ or Bun, SQLite 3

1. Clone & Build the Backend

git clone https://github.com/loocor/MCPMate.git
cd MCPMate/backend
cargo build --release

2. Start the Proxy

cargo run --release

The proxy starts with:

• REST API on http://localhost:8080
• MCP endpoint on http://localhost:8000

3. Launch the Dashboard

cd ../board
bun install
bun run dev

Dashboard available at http://localhost:5173.

Coming soon. An online environment will let you explore the dashboard, profiles, and client configuration without a local setup.

• Integrated mode (desktop) — Tauri bundles backend + dashboard for local all-in-one operation
• Separated mode (core server + UI) — Run backend independently and connect either the web dashboard or desktop shell to that core service
• Client mode flexibility — Managed clients can continue using hosted/transparent workflows while the control plane runs remotely

# Run all checks
./scripts/check

# Start backend + board together
./scripts/dev-all

See AGENTS.md for development guidelines, coding standards, and contribution workflow.

1. Account-based configuration backup & restore
2. Skills-mode packaged profiles
3. Downstream MCP client OAuth authorization management
4. Cross-platform release readiness — desktop OS stability, container-based deployment, and Homebrew installation support

Contributions are welcome! Please:

1. Read AGENTS.md for development guidelines
2. Open an issue to discuss significant changes
3. Submit pull requests against the main branch

GNU Affero General Public License v3.0 (AGPL-3.0)

Built with ❤️ by Loocor