I build AI agent tooling, MCP servers, and evaluation environments.
Most of my day-to-day work lives in private product repos. Publicly, the clearest slice of my work is the tooling and infrastructure I have been building around agent workflows, local integrations, and developer systems. I use coding agents heavily and care a lot about where they fail, why they fail, and how to build cleaner interfaces around them.
- AI agent tooling and MCP servers
- Evaluation harnesses and model failure analysis
- Local-first automation for real workflows
- Swift and Apple-platform developer tooling
- safari-mcp - Local Safari control, extraction, and screenshots for agent workflows
- multimodal-imessage-mcp - iMessage access, conversation search, and attachment handling
- app-store-connect-mcp - App Store Connect automation and release operations
- gmail-multi-inbox-mcp - Multi-account Gmail integration with built-in OAuth onboarding
- codex-memory - CLI-first codebase memory for risk, co-change, and decision context
- codex-sessions - Small CLI for seeing which Codex sessions are live on your machine
- sourcekit-lsp-marketplace - Swift and Objective-C support for Claude Code via SourceKit-LSP
TypeScript, JavaScript, Python, Go, Swift, SwiftUI, SQL, APIs, Railway, Xcode.
I like simple systems, root-cause debugging, and tooling that has to hold up in real workflows.
- GitHub: @tszaks
- Email: [email protected]