MCP Gateway (Lazy Router)

January 22, 2026 · View on GitHub

MCP Manager Icon

The "Lazy Loading Router" for MCP tools.

This extension has been refactored from a UI-based dashboard into a high-performance, background service that orchestrates Model Context Protocol (MCP) servers for VS Code and Antigravity.

🚀 How It Works

Instead of running all your MCP servers simultaneously (wasting RAM/CPU), MCP Manager acts as a router:

  1. Orchestration: It uses uv (a super-fast Python package manager) to manage server environments.
  2. Lazy Loading: Tools are only spun up when an AI agent explicitly requests them.
  3. Bridge: It bridges your existing configuration into the VS Code Language Model API.

⚡ Features

  • Auto-Bootstrap: Automatically installs and manages uv.
  • Zero-Config Migration: Migrates your existing mcp-config.json to a router_manifest.json.
  • Antigravity Sync: keeps your MCP configuration synchronized with the Antigravity agent.

📦 Usage

The extension works automatically in the background.

  1. Install: Sideload the .vsix.
  2. Sync: The extension activates on startup (onStartupFinished).
  3. Verify: Run the command MCP: Show Status to see the list of active tools and the router status.

🔧 Configuration

The extension manages its own router_manifest.json.

  • Commands:
    • MCP: Sync Configuration: Force a sync of the router manifest.
    • MCP: Show Status: detailed status notification.

🧪 Verification & Usage Walkthrough

Follow these steps to verify that MCP Gateway is correctly orchestrating your tools for your AI Agent.

Step 1: Open the Dashboard

  1. Press Ctrl+Shift+P (Cmd+Shift+P on Mac).
  2. Type and select MCP: Show Status.
  3. A premium MCP Manager Dashboard will open.
    • check for the "Connected" badge.
    • Note the Active Servers count.

Step 2: Ask your Agent

  1. Open GitHub Copilot Chat (or your Antigravity Agent).
  2. Ask a question that requires a tool.
    • Example:

    "Please search online for the current time in Tokyo using Brave Search." (Note: Use MCP: Add Tool to install usage-ready tools first).

Step 3: Watch it Happen (Accountability)

  1. The Agent will mistakenly or correctly identify the tool and send a request.
  2. MCP Gateway intercepts the request, spins up the Python environment via uv, executes the tool, and shuts it down.
  3. Switch back to the MCP Manager Dashboard.
  4. Look at the Accountability & Recent Logs table.
    • You will see a new entry: User (Default) | brave_search | Success.
    • The Usage Share bar for that tool will update.
    • This proves the end-to-end connection is working!

3. Adding New Servers (One-Click)

Adding new tools is now fully automated.

  1. Press Ctrl+Shift+P.
  2. Select MCP: Add Tool.
  3. Choose a tool from the Premium Catalog (SQLite, Brave Search, Fetch, Git, etc.).
  4. The extension will automatically:
    • Ask for necessary API Keys or Paths.
    • Configure the environment.
    • Sync with the Agent.

Note: Manual JSON editing is no longer required.

🤝 Contributing

Powered by Craze for AI and Model Context Protocol (and lots of love for coding)