Claude Code and MCP workflows¶
Claude Code is the primary execution surface for FA work. This page is the practical setup and usage reference. For where it fits in your first days, see the FA onboarding guide.
What Claude Code is¶
Anthropic's agentic command-line tool. You run it in a terminal inside a project folder, describe a task in plain English, and it reads your files, writes and edits code, runs commands, and works the repo — pausing for you to review each change. It is how non-engineering work gets executed at SCAILE.
It's also how you answer questions from the org: point it at any repo in the scailetech org — this playbook, the knowledge base, a venture, a client — and ask in plain language. See Query in Claude Code.
Install¶
Claude Code ships as a native installer — there is no npm package, and Claude Code itself does not need Node.js.
irm https://claude.ai/install.ps1 | iex
curl -fsSL https://claude.ai/install.sh | bash
brew install --cask claude-code
Native installs auto-update. A venture's codebase will still have its own toolchain (Node.js, Git, etc.) — install that per the project README.
Plan and login¶
You need a paid Claude plan that includes Claude Code. At SCAILE that's the company Team plan — sign in with the company account on first run, not a personal one. Don't expense a personal Pro plan without checking first (the exact plan + who adds you is an open question for your guide).
cd path/to/project
claude # first run opens the browser to authenticate
/login switches accounts inside a session.
Give it context¶
Claude Code reads project files on its own. The file that matters most is CLAUDE.md in the repo root — persistent instructions loaded at the start of every session (build commands, conventions, do's and don'ts). Run /init to have Claude draft one from the codebase, then edit it. This playbook repo's own CLAUDE.md is what lets it answer questions about itself — see use-in-claude-code.md.
Everyday commands¶
| Command | What it does |
|---|---|
claude |
Start an interactive session in the current folder |
claude "do X" |
One-off task from the shell |
claude -c |
Continue the most recent conversation here |
/init |
Create or improve the repo's CLAUDE.md |
/help |
List every command |
/clear |
Reset the conversation when you switch tasks |
/context |
See what's filling the context window |
Shift+Tab |
Cycle permission modes (Default → auto-accept → plan mode) |
Esc |
Stop Claude mid-turn |
Daily rhythm: pick one concrete outcome → run claude → for anything non-trivial, ask for a plan first (plan mode) and read it → review every diff → commit in small steps → push to a branch/preview, never production without sign-off. Full version in the daily workflow.
MCPs¶
MCP (Model Context Protocol) servers connect your tools directly to Claude Code. Configure these once: Notion, Gmail, Drive, Slack, Granola. Add each server to your Claude Code MCP config; after that they're available from any project. You can also point Claude Code at this playbook — or any repo in the scailetech org (the knowledge base, a venture, a client) and query it directly, with or without MCP — see Query in Claude Code.
The shared library is still being built¶
There is no standard SCAILE library of Claude Code skills, prompts, and MCP setups yet — everyone improvises. Building that shared library is an early FA contribution (the confirmed gap in overview.md). As you find prompts, skills, and MCP configs that work, capture them here and in the ideas log so the next FA doesn't start from zero.
Open questions¶
| Question | Owner | Urgency |
|---|---|---|
| Is Claude Code fluency a prerequisite or a ramp-up? | Simon | Medium |
| Which Claude plan does the FA use, and who adds you to it? | Simon / Niccolo | Medium |
Owner: Current FA Last reviewed: 2026-07-03