Three ways to get Claude Code configured for a real repo. One of them is free and is the right answer for a lot of people, so it goes first.
Last updated: 2026-09-21
| Do it yourself | Setup Lite | Setup Sprint | |
|---|---|---|---|
| Price | Free, MIT | $29 one time | $99 one time |
| What you get | A worked CLAUDE.md, a permission allowlist with both an allow and a deny list, three skills and a formatter hook — generic, for you to fill in | A CLAUDE.md with your commands, the allowlist built from what your repo actually runs, and one skill | All of Lite, plus hooks, 3–5 skills, up to 3 MCP servers wired, a one-page runbook and an ACCEPTANCE.md |
| How it arrives | You clone a public repo and copy files in | A pull request you read and merge. Never a push to your default branch | Same: a pull request you read and merge |
| Your time | An afternoon of reading, per repo | Three form fields, then reviewing a PR | Same three fields, plus naming the MCP servers you want |
| Clock | Now | 24 hours from a complete intake | 48 hours from a complete intake |
| Who does the reading | You | We do, from your code. Every command written down was run in your repo first | Same, across more surface |
Clocks run from a complete intake, not from the Stripe receipt. “Not sure” is a complete answer to any intake field — working the rest out of your code is the job you are paying for, not a prerequisite.
Also free and not gated behind an email: a full audit we published of our own starter repo, and a guide to writing skills that gives away the part people get wrong.
What it is not: not hooks, not MCP servers, not 3–5 skills. If you already know you want those, skip a rung — a scope change is a new order, not a revision round.
Setup Lite $29You already have a setup and it is fighting you. Adding another setup on top usually creates a second source of truth instead of removing the first. That is the Claude Code Surface Audit at $79: a written audit of the CLAUDE.md and settings you already have, a contradiction map, an exact fix list, 48 hours.
You do not have a repo yet. Then none of this is the right buy and we will say so rather than stretch it. Take the free starter, build something with it, and come back when there is a repo to set up.
You want a tool built rather than a repo configured. That is a different line of work: MCP Basic $199 for one tool, or Build Packet $399 for a server or a Worker. Configuration will not get you there. Side by side: Setup Sprint vs MCP Basic.
You are not sure Claude Code is for you at all. Read the plain-language guide first. It costs nothing and it talks a fair number of people out of buying, which is the point.
You should not, if you enjoy the reading. The free starter is a complete, working setup and it is generic on purpose — you fill in your own commands, your own traps, your own skill. The $29 buys the reading instead: a CLAUDE.md written from your actual repo, an allowlist built from the commands it actually runs, and every one of those commands run in your repo and passed before the pull request opens. That check is the $29.
Lite is one repo, one skill, the allowlist, 24 hours. Sprint is 3–5 skills, hooks, up to 3 MCP servers wired, a one-page runbook and an ACCEPTANCE.md, 48 hours. Rule of thumb: if the complaint is “it does not know my repo”, buy Lite. If it is “it asks before every command and nothing runs after it edits”, that is Sprint.
An afternoon for the first repo, honestly — and far less for the second, because by then you have made the decisions. Copying the files takes minutes. The time goes on deciding which commands are safe to allowlist, which annoyance deserves a skill, and discovering that a command you wrote down does not run from a clean checkout.
Yes, and nothing expires — neither rung is a trial. The honest caveat: a scope change is a new order rather than a revision round, so if you already know you want hooks or an MCP server, buying Sprint once costs less than buying both.
Collaborator access to a branch, and nothing else. The deliverable is a pull request you review, never a push to your default branch. We never ask for an API key or a .env, and if one arrives by accident we tell you to rotate it and keep it out of the work.
Both, and we would rather say so. AI tools including Claude produce the work, and a person reads every deliverable before it ships — including checking that every command written into the CLAUDE.md was run in your repo and passed.
This page names a price sixteen times and until now gave you nowhere to act on it. Here are all three, in the order they are argued above — nothing below is a trial and nothing expires.
Doing it yourself costs an afternoon and $0: kit-claude-code-starter is a complete working setup, MIT, no email wall. If the reading sounds like a good evening, start there and close this tab — that is a real answer, not a consolation one.
Buy Setup Lite — $29 Buy Setup Sprint — $99 See every price and scopeBuying the wrong rung is recoverable and we would rather say how: a scope change is a new order, not a revision round, so if you already know you want hooks or an MCP server wired, buying Sprint once costs less than buying Lite and then Sprint.