VS Code MCP Supergateway

A centralized, multi-client Model Context Protocol (MCP) gateway that aggregates, routes, and offloads context between IDE clients, local LLM workers, and backend MCP services.

💡 Overview & Value Proposition

As MCP usage grows, managing multiple disjointed MCP servers across different clients (VS Code/Copilot, LM Studio, etc.) becomes cumbersome. VS Code MCP Supergateway acts as a unified hub:

1. Multiplexing & Routing: Connect multiple clients to a single, orchestrated MCP endpoint.

2. Context & Token Savings: Keep expensive Frontier Model context clean by delegating repetitive context-structuring tasks to local worker models.

3. LM Studio Loopback Pattern: Turn LM Studio into both an MCP consumer and a high-speed local MCP tool/worker for tasks like documentation linting, diff summarization, and task management.

Related MCP server: MCP Gateway

🏗 Integration Flow

graph
    subgraph Clients["Clients"]
        Copilot["VS Code / Copilot Chat<br/>(Cloud Frontier Models)"]
        LMStudio_Host["LM Studio Client UI"]
    end

    subgraph Gateway["HTTP MCP Gateway"]
        Supergateway["Supergateway Proxy"]
    end

    subgraph Loopback["Local Worker Loopback"]
        LMStudio_Worker["LM Studio MCP Server<br>(Local Small Models: Gemma, LFM, Qwen, ...)"]
    end

    subgraph Backends["Local MCP Servers"]
        CodebaseMemory["Codebase Memory"]
        MarkdownVault["Markdown Vault<br>(Tasks, ADRs & Specs)"]
        GitLab["GitLab Integration"]
    end

    %% Client connections to Gateway
    Supergateway <-- "Local Memory & Mgmt Tools" --> Clients
    
    %% Gateway routes to Backends and Loopback Worker
    Supergateway --> CodebaseMemory
    Supergateway --> MarkdownVault
    Supergateway --> GitLab
    
    %% Loopback Worker Execution
    Supergateway <-->|"Local Sub-Agents"| LMStudio_Worker
    LMStudio_Worker -. "Reads/Updates" .-> MarkdownVault

⚡ Key Features & Concepts

• Unified Control Plane: Connect Copilot and LM Studio simultaneously to your underlying toolchain (GitLab, Codebase Memory, Vault).

• Sub-Agent Loopback: Offload context aggregation, diff generation, and documentation updates to fast local models running in LM Studio without consuming cloud tokens.

• Markdown Vault Integration: Structure project tasks, architectural decision records (ADRs), and feature specs directly in markdown files guarded by YAML access controls (agent_access).

🚀 Getting Started

Prerequisites

• Node.js 20+

• VS Code with GitHub Copilot

• LM Studio (optional, for local model offloading & loopback)

Installation

git clone https://github.com/Flexi23/vscode-mcp-supergateway.git
cd vscode-mcp-supergateway
npm install
npm run build

🗺 Roadmap & Future Plan

Phase 1: MVP & Core Gateway (Current)

• Basic stdio / SSE transport routing.

• Multi-backend server orchestration (Codebase Memory, GitLab, Vault).

• Initial agent-assisted development groundwork.

Phase 2: LM Studio Loopback & Context Worker

• Implement local LM Studio MCP tool wrapper (summarize_diff, generate_adr).

• Add zero-blocking async tool handling for local inference.

• Graceful fallback & timeout management when local GPUs are under heavy load.

Phase 3: Vault & Task Management Enhancements

• Standardized YAML-Frontmatter parser for Markdown Vault (tasks/, adrs/).

• Agent scope security layer (agent_access: read | append | edit | hidden).

• Automated context bundle generator for Copilot prompts.

📄 License

MIT License. Feel free to contribute or adapt!