docker/mcp-gateway

What it solves

This project provides a unified way to manage and deploy Model Context Protocol (MCP) servers. It removes the friction of manually managing the lifecycle of multiple isolated MCP servers, handling authentication (secrets and OAuth), and ensuring consistent tool discovery across different AI clients like VS Code, Cursor, and Claude Desktop.

How it works

It implements a gateway pattern where an AI client connects to a single MCP Gateway, which then routes requests to various MCP servers running as isolated Docker containers. The tool is provided as a Docker CLI plugin (docker mcp) that allows users to organize servers into "profiles" (groups of related servers) and manage them via OCI-based catalogs.

Who it’s for

Developers who use MCP-compatible AI clients and want a secure, containerized way to manage multiple external data sources and tools for their LLMs without manually configuring each client.

Highlights

• Container-based Isolation: Runs MCP servers in Docker containers to ensure security and isolation.
• Unified Interface: Provides a single gateway for AI models to access multiple servers.
• Secrets & OAuth: Integrates with Docker Desktop's secrets management and provides built-in OAuth flows for service authentication.
• Profile Management: Allows grouping, exporting, and sharing server configurations via OCI registries.
• Dynamic Discovery: Automatically discovers tools, prompts, and resources from running servers.