A power user's 5-layer Claude Code setup — CLAUDE.md, MCP, hooks, and subagents explained
A developer shared their full Claude Code configuration that turns the tool into an autonomous, around-the-clock system. The setup has five layers: project rules in CLAUDE.md, external tool connections via MCP servers, reusable workflows as skills, automated safety scripts as hooks, and parallel workers as subagents. The author cut an 8-hour blockchain analysis task down to 15 minutes.
The CLAUDE.md file acts as a 'project operating system' — it holds concrete rules like file length limits, voice note routing, and task extraction steps so Claude behaves consistently every session without being reminded. MCP servers plug in external services: Coolify for deployment automation, Telegram for reading and sending messages, and Codex for dual-model code review, letting Claude directly control those tools.
Hooks are small shell scripts that fire automatically on specific events — for example, a pre-commit hook blocks any file ending in .env or .key from being accidentally committed, making unattended overnight runs safe. Skills are markdown-written reusable procedures (no code compilation needed) for tasks like writing SQL queries or cross-posting in multiple languages. Subagents run in parallel — one fetches data, one profiles wallets, one writes the report — all at the same time. The author values Claude Max at $100/month because it saves roughly two hours of work per day.
Key points
- Put only always-true rules in CLAUDE.md; move occasional procedures into skills instead
- MCP servers let Claude directly operate external tools like deployment platforms and messaging apps
- A simple hook script can block accidental commits of .env or secret key files — critical for safe unattended use
- Three subagents working in parallel can collapse an 8-hour research task into 15 minutes
- A full voice-to-completion pipeline is possible: voice note → transcription → Claude executes the task automatically
Quick term guide
- autonomous
- The ability of an AI to complete tasks or make decisions without constant human guidance.
- MCP servers
- Servers that help an AI tool connect to outside services or company data.
- MCP server
- A server that helps AI tools connect to outside services in a standard way.
- blockchain
- A digital record-keeping system where data is stored across many computers instead of one central place.
- deployment
- The process of putting software changes into a running system.
- automation
- A way to make repeated work happen without doing every step by hand.
- code review
- A check of code before it is shipped, usually to find mistakes or improvements.
- transcription
- Automatically converting spoken words into a written text file during a meeting.