Project configuration

Complete project setup layers

Five layers that turn Claude from a generic assistant into a project-aware team member.

The five layers

Layer 1context

CLAUDE.md instructions

A markdown file that tells Claude about your stack, conventions, and constraints. It is loaded automatically at the start of every session.

~/.claude/CLAUDE.md

/project/CLAUDE.md

Layer 2capability

Skills library

Reusable prompt-and-tool bundles that teach Claude specific workflows — testing, deployment, database migrations, and more.

.claude/skills/

~/.claude/skills/

Layer 3roles

Agent definitions

Named agent profiles that combine a persona, a set of skills, and permission boundaries. Each agent focuses on one job.

.claude/agents/

~/.claude/agents/

Layer 4access

MCP server connections

Model Context Protocol servers that give Claude live access to databases, APIs, file systems, and other external services.

/project/.mcp.json

Layer 5extend

Community skills + marketplace plugins

Third-party skills and plugins from the community marketplace. Drop them in and Claude gains new capabilities instantly.

github.com / marketplace

What your AI can reach

filesystem

Your project files

Read, create, edit, and delete files anywhere in your project tree. Claude sees the same directory you do.

database

Your live data

Query and modify your database through MCP connections — Postgres, SQLite, Supabase, and more.

github

Your code history

Browse commits, create branches, open pull requests, and review diffs without leaving the terminal.

memory

Session scratchpad

Claude can write and read temporary notes during a session to track decisions, plans, and intermediate results.

web search

Live internet data

Search the web for documentation, Stack Overflow answers, and package information in real time.

deployment

Your live environment

Trigger deploys, check build status, and tail logs from Vercel, Railway, or your own CI pipeline.

Where your data lives

Database

Supabase, Postgres, SQLite, or any database reachable through an MCP server or CLI.

File storage

Local disk, S3 buckets, or cloud storage accessed via signed URLs and CLI tools.

Deployment

Vercel, Railway, Fly.io — wherever your app is hosted and your builds run.

GitHub

Repositories, issues, pull requests, Actions workflows, and the GitHub API.

API services

Stripe, Resend, OpenAI, and any third-party service your project depends on.

Local machine

Your terminal, file system, running processes, and environment variables.