MCP‑Collection

What is MCP‑Collection?

A single Docker image that bundles multiple Model‑Context‑Protocol (MCP) servers and exposes them through a unified entry‑point. It is designed for Kubernetes deployments and provides remote access out of the box via the supergateway bridge.

How to use MCP‑Collection?

Pull the multi‑arch image from GitHub Container Registry, then run it with the name of the desired MCP server as an argument. To list available servers, run the image without arguments. For example: docker run --rm ghcr.io/erhardtconsulting/mcp‑collection:1.0.0 mcp-server-fetch.

Key features of MCP‑Collection

• One image contains many MCP servers, chosen at runtime.
• Remote‑ready via supergateway (HTTP/SSE or WebSocket).
• Kubernetes‑first with health endpoints and non‑root user.
• Deterministic builds: all dependencies locked, no runtime installation.
• Image size ≤80 MB using a distroless runtime layer.

Use cases of MCP‑Collection

• Deploy multiple MCP servers in a Kubernetes cluster from a single image.
• Expose an MCP server over HTTP/SSE for remote AI agents.
• Run a private MCP server for fetching web pages (mcp-server-fetch).
• Integrate a Qdrant vector‑search MCP server into a containerized pipeline.

FAQ from MCP‑Collection

Which MCP servers are included?

The image ships mcp-maven-deps, mcp-searxng, mcp-server-calculator, mcp-server-fetch, mcp-server-puppeteer, mcp-server-qdrant, mcp-server-time, mcp-yahoo-finance, and npm-search-mcp-server.

How does the entry‑point select and run a server?

The entry‑point activates a virtual environment, validates the server name against a fixed whitelist, and executes supergateway --stdio "<server> [extra‑args]" to bridge the server’s stdio interface to HTTP/SSE.

Is the container secure?

The container runs as a non‑root UID (1000:1000), removes the shell after exec, and recommends using a reverse proxy for TLS and authentication.

How can I add a new server to MCP‑Collection?

Fork the repository, add the server dependency to package.json or pyproject.toml and update entrypoint.sh, run make test, and open a pull request.

What are the runtime requirements?

Only Docker is required; Kubernetes is optional. The image is multi‑arch and uses a distroless base.