Notes from building software solo — MCP servers, AI coding agents, and the APIs underneath them. Everything here comes out of something I actually shipped and had to debug, so the numbers are real and the parts that did not work are in there too.
Follow new posts by RSS.
I have dictated 147,177 words with Wispr Flow since January 2025, most of them into Claude Code. My real numbers, the price, and what goes wrong.
· wispr flow, dictation, claude code, tool review
Claude Code reviews code four ways: /code-review, ultrareview, managed Code Review and GitHub Actions. What each one does, and the review rules I use.
· claude code, code review, ai coding agents
How to build an MCP server with the official laravel/mcp package, and six problems I had to fix on my three production Laravel MCP servers.
· laravel, mcp, ai agents
I use Claude Code and Codex daily on production Laravel and TypeScript work. Where each one wins, where benchmarks stop helping, and my workflow.
· claude code, codex, ai coding agents
The console path to an xAI API key, why the OpenAI SDK works unchanged, and a working Telegram bot — plus the billing detail that catches people out.
· grok, xai, api, bots
Build a production MCP server: tool design, transports, OAuth, and the five failures that only show up once a real model starts calling it.
· mcp, tutorial, ai agents, laravel
MCP does not replace your API. It sits on top of one. What changes when you add an MCP server, what does not, and when building one is a waste of time.
· mcp, api, ai agents