MCP Server Collection
← Project Index MCP Server Collection

MCP Server Collection

A grab-bag of Model Context Protocol servers wiring assistants into the things I actually use -; from live EEG streams to an OAuth-authenticated bridge into a school communication system.

Archived Started: Summer 2025 Updated: Summer 2025

Overview

Once you have one MCP server, you have a habit. This is the collection that habit produced -; a handful of servers, each connecting an assistant to a different data source or tool. They range from small utilities (read my Obsidian vault, query a SQL Server, capture audio) to two genuinely interesting ones: a live EEG bridge and an authenticated gateway into a school's app.

Background

MCP turned out to be the right shape for a lot of "I wish the assistant could just see X" problems, and the marginal cost of a new server is low once you've written a few. So they accumulated -; some Python, some TypeScript -; until the collection itself was worth cataloguing.

The Servers

The collection, roughly in order of ambition:

  • Obsidian MCP (Python) -; read/write/organize markdown across vaults, with wikilink and YAML-frontmatter parsing and fuzzy note resolution.
  • SQL Server MCP (Python, pyodbc) -; parameterized queries, schema introspection, transactions, plus an optional natural-language-to-SQL module.
  • Audio Capture MCP (Python, sounddevice) -; microphone capture with FFT analysis, RMS/peak metrics, and silence detection.
  • MindWave EEG MCP (Python) -; streams EEG from a NeuroSky MindWave headset over Bluetooth (ThinkGear, 512 Hz) and does 8-band frequency analysis -; delta, theta, alpha, beta, gamma -; plus attention/meditation metrics and cognitive-state detection.
  • NHA App MCP (TypeScript/Node.js) -; an Okta OAuth2 bridge into the NHA school communication app, covering posts, chat, directory, calendar, groups, notifications, and schools, with automatic token refresh.

The EEG server is the one that feels like science fiction: a sliding FFT over the raw 512 Hz stream produces band powers, computed the textbook way after a Hamming window:

band power = Σ |FFT(x · w)|² over the band -; e.g. alpha = 8-;13 Hz (w = Hamming window)

The NHA App server is the most production-shaped: it does the full client-credentials / refresh-token OAuth dance against Okta, re-authenticates automatically when a token expires, and exposes seven API surfaces of the school app through a single MCP interface.

Current Status

Archived. The servers work individually; this card is the catalogue rather than a single product. One member of the collection -; an early geospatial experiment -; outgrew the others and became its own thing.

  • Five-plus working servers across Python and TypeScript.
  • EEG streaming with 8-band frequency analysis.
  • NHA App bridge with Okta OAuth and token refresh.