Labsco
MCP SERVER

MCP Workshop Servers

by slcc2c

Ten local servers — files, Docker, memory, Postgres, Redis, Mongo, Neo4j, Jupyter, Kubernetes, Xcode — started together behind one gateway.

MCP Plumbing: Proxies, Registries & Inspectors
Summary
Ten servers on one switch, for the machine you experiment on.

The value is not any single one of the ten — it is having files, containers, four databases, notebooks and a memory graph reachable from the same session, so a throwaway experiment does not begin with an hour of wiring. It is one person's workshop rig rather than a distributed package: you clone it, enable the servers you want, and keep the secrets in 1Password instead of the config file.

What it is

One gateway process that starts ten MCP servers written for this project and routes each call to the right one: filesystem, Docker, an in-memory knowledge graph, PostgreSQL, Redis, MongoDB, Neo4j, Jupyter, Kubernetes and Xcode builds. Secrets are not pasted into the configuration — connection strings and tokens are written as 1Password references and resolved at start. The memory server segregates entries by project tag, so several experiments can share one instance without reading each other's notes. The GitHub tier is the exception: it runs the published `@modelcontextprotocol/server-github` package rather than code from this repository.

What you get
  • Sandboxed file work: read, write, copy, move and delete, directories listed and created, stats and existence checks, search, and watchers on a path
  • A knowledge graph an assistant writes to and searches, tagged per project, with related entries and an export
  • Docker containers created, started, stopped, inspected and pruned, images pulled and built, networks and volumes managed, logs and stats read
  • PostgreSQL queries and transactions, tables listed and inspected, and select, insert, update and delete without hand-writing SQL
  • Redis across its data types with pub/sub and pipelines, MongoDB find, aggregate and index work, and Neo4j Cypher with path finding and graph algorithms
  • Jupyter notebooks created, opened and executed cell by cell or whole, with variables inspected, plots produced and export to other formats
  • Kubernetes namespaces, pods, deployments, services, configmaps, secrets and jobs, with pod logs and cluster events
  • Xcode projects, schemes and targets listed, Swift package dependencies managed, and iOS simulators created, booted and loaded with an app
Requirements

Node and a clone of the repository — it is not published to a package registry, so the gateway runs from the build in your checkout. A GitHub token goes in the environment for the GitHub tier. Everything else is opt-in per server: the database servers need credentials only for the databases you actually enable, and the project expects those as 1Password references rather than literals. The Xcode server is macOS only.