I build AI agents and MCP tools — and the Claude Code skills, serverless backends, and native apps around them. Mostly things I actually use.
🧰 Featured — claude-code-toolkit
A plugin of Claude Code skills + a failure-mode hook, for building MCP servers well and catching bugs before they ship. The headline pair:
/scaffold-mcpgenerates a runnable, tested, ruff-clean MCP server in the pure-core + thin-adapters shape (FastMCP, dual stdio/HTTP, CLI, tests, bundled seed) — and can stand up a full worked example to drive live./verify-mcpproves one is green: venv + ruff + pytest + a smoke import- a
@mcp.tool()inventory → a GREEN/RED health report. Scaffold builds it, verify proves it.
- a
Plus public-ready, design-note, handoff, note, failure-scan, and a
PostToolUse hook that auto-flags risky patterns (e.g. unpaginated DynamoDB scans).
- meal-planner-mcp — plans a week of meals from a local recipe library: optimizes ingredient overlap, reuses leftovers, avoids recent repeats, then emits a shopping list + Markdown plan. Self-contained, clone-and-run.
- mood-mixer — mood-based Spotify playlist creator. Rebuilds the audio features Spotify deprecated (Nov 2024) from community data (AcousticBrainz / GetSongBPM).
- MagicMonitor — live Walt Disney World ride-status dashboard + Pushover alerts (Lambda · DynamoDB · EventBridge · Amplify), with an MCP-exposed agentic trip planner.
- decades-trivia — adaptive decade-based trivia on AWS (Bedrock · Lambda · DynamoDB). Questions generated by Claude, fact-checked by two-model consensus, with Wikipedia RAG behind review-facts.
- python-turtle-workshop — zero-install, browser-based workshop that teaches kids real Python with a turtle and an optional AI helper. Runs entirely in the browser via Pyodide.
- FrameShuffle — rotate your own art & photos on a Samsung Frame TV from a local web UI: shuffle, folder / color filtering, and battle-tested 2024-model quirk handling. No cloud, no account.
- TripSplit — native iOS app (SwiftUI) for splitting shared trip expenses: flexible per-item splitting and a graph-based algorithm that settles the group in the fewest payments.
Python · TypeScript · Swift · AWS (Lambda, DynamoDB, EventBridge, Bedrock, Amplify) · Next.js · FastMCP · Claude Code skills & plugins