AI agent tooling · developer tools · TypeScript · Python · Java
I build practical tools for developers and AI agents. My recent work sits where local software, protocol interoperability, and coding workflows meet.
- Building local-first tools that make coding agents more capable and easier to operate.
- Exploring MCP, multi-protocol LLM gateways, agent workflows, and developer experience.
- Turning experiments into installable tools with clear documentation and repeatable workflows.
|
A local LLM proxy with Anthropic, OpenAI, and Responses API routing, protocol translation, streaming conversion, token tracking, and a macOS app. TypeScript macOS LLM APIs
|
A composable extension collection for the Pi coding agent, including output distillation, tool supervision, metrics, model discovery, and workflow orchestration. TypeScript Pi Agent tooling
|
|
A CLI for managing MCP servers, discovering tools, executing calls, and keeping persistent connections through daemon mode. TypeScript MCP CLI
|
Start with a real problem
Make the smallest useful tool
Keep the interfaces explicit
Document the path from install to first result
Ship, observe, and improve
Before working on AI and developer tooling, I built Java services, web applications, and small games. The projects are different, but the thread is the same: learn by building systems end to end.
- JavaGame, a collection of Java games including 2048, Snake, Tetris, and Plane War.
- bitbyte, a Java Spring Cloud backend for a computer science blog.
- homebrew-tap, Homebrew distribution for my tools.
- GitHub: github.com/maplezzk
- Email: zhangzikuan4513@gmail.com
Build useful things. Keep the edges sharp.