I recently built a full Expenses feature for my invoicing app — creation, editing, dashboard totals, and export — in about 15 minutes.
And I did it using Claude Skills as MCP tools, completely outside Claude.
It all started after seeing a post from Matt Shumer, where he shared a simple two-step framework for handling complex tasks:
- Analyze the problem
- Create a plan
I used that process many times, but constantly copy-pasting the prompts was annoying. When Claude Skills launched, I decided to turn this workflow into a reusable Skill — a small, self-contained automation I could trigger anytime.
Then I took it a step further.
Since I already had MCPBundler, I integrated Claude Skills into it — so now Skills can run as MCP tools. This means I can call any Claude Skill directly from Codex or any other MCP-compatible AI tool, without ever opening Claude itself.
Using that setup, I built the Expenses module for my invoicing app almost one-shot:
- Add and edit expenses (per customer or project)
- Attach invoices
- Filter and export all data
- View total spending on the dashboard
It came together fast, clean, and functional — exactly how AI-assisted workflows should feel.
Running Claude Skills as MCP tools turns them into something much more powerful: part of your actual dev environment, not just a chat. It’s reusable, automated, and practical.