About PromptWritingStudio
A working resource for people building real work with Claude — Claude Code, MCP, sub-agents, and hooks.
Run by Bryan Collins, writer and builder at Become a Writer Today.
What this site is
PromptWritingStudio started as a prompt-engineering resource back when people were still pasting snippets into ChatGPT. It has narrowed since.
Today it focuses on Claude — specifically the parts of the Claude stack that most material on the web glosses over: Claude Code, MCP servers, sub-agents, skills, hooks, CLAUDE.md files, and the plan-vs-API decisions that actually show up when you start shipping real work.
Every guide here is pulled from real sessions. The MCP recipes are servers I run. The CLAUDE.md playbook is the one I use on this repo. The pricing and model data is pulled from a single source and checked weekly by an automated drift-check job — so when Anthropic ships a new model, the calculators and comparisons catch up within a week.
If you are a solo operator, small team, or in-house AI lead trying to get Claude Code into production without burning a week on config, this site is for you.
Who runs it
I'm Bryan Collins. I write books and essays at Become a Writer Today and publish AI Flash Report, a daily AI-news briefing.
I use Claude Code every day to build and maintain a small portfolio of sites — this one, a few news aggregators, a B2B directory, and some tools. Most of what you'll read here is the stuff I figured out while shipping that work, written up so the next person doesn't have to spend the same afternoon on it.
I'm not an Anthropic employee. This is an independent site. Where I link to third-party tools, assume it may be an affiliate link unless stated otherwise — see the affiliate disclosure for details.
How we keep content accurate
Single source of truth
Claude model IDs, prices, context windows, and plan limits live in one JSON file. Every calculator and comparison page reads from it. Change the file, every page updates.
Weekly drift check
A scheduled job fetches Anthropic's models page and pricing page every Monday and opens a GitHub issue when anything has changed since we last verified.
Verified stamps
Pages with volatile data (prices, plan limits) carry a "Last verified" stamp linking to the source. If a stamp is old, the number is probably too.
Working guides only
If I haven't actually used it, I don't publish it. The MCP stack, hooks recipes, and CLAUDE.md playbook are all from real projects — not speculation.
Spotted something wrong?
Every page has a last-verified stamp. If you hit something that's out of date or plain wrong, tell me and I'll fix it.
Get in touch