This is the fastest low-risk path for a new teammate.
Goal:
- get Claude Code running in one repo
- load a shared starter baseline
- connect a small MCP set
- confirm the workflow feels predictable before adding more power
From this repository:
node scripts/install.mjs /path/to/your-projectUse the base starter first. Do not add variants on the first pass unless the repo already has a clear team convention for them.
If the repo obviously fits one of the bundle presets, start from docs/bundle-selector.md or docs/examples-gallery.md instead of building the install from scratch.
For most teams, begin with:
- GitHub
- docs or knowledge base
- tasks or ticket tracking
The smallest ready-made profile is:
mcp/examples/team-onboarding.json
Copy only the servers you actually need into your project .mcp.json.
Start Claude Code in the target project:
cd /path/to/your-project
claudeThen run:
/help
/check-setup
/agents
/mcp
What you want to see:
- the project commands are visible
- the setup audit does not show obvious gaps
- the shared agents are present
- the MCP servers you added can be discovered
Review these files before broader team use:
.claude/settings.jsonCLAUDE.md.mcp.json
If you are unsure where a change belongs, use config-layers.md before editing.
Usually the first changes are:
- tighten or relax permissions for the repo
- replace placeholder team guidance in
CLAUDE.md - wire real environment variables for MCP access
Use a tiny change or sample diff and run:
/review
Then try:
/spec
This confirms the shared workflow is understandable before anyone depends on it for real delivery work.
Good next steps:
frontendfor UI-heavy reposbackendfor service repossecurityfor higher-risk systemsrelease-engineeringfor careful rollout workflows
If you combine variants, later variants win when they write the same file path.
Day 1:
- base starter
- onboarding MCP profile
/check-setup
Day 2:
- one harmless review task
- one harmless planning task
Day 3+:
- add one variant if the repo clearly needs it
- enable one small hook recipe if the repo needs tighter guardrails
- add operational MCP servers only after the baseline works