These two get confused constantly, and the confusion costs real money. One configures what Claude Code already has. The other builds it something it does not have.
Last updated: 2026-09-21
If you are unsure, it is almost always the first one. Most problems that feel like a missing tool are a missing allowlist entry or a CLAUDE.md that never said which command to run.
| Setup Sprint — $99 | MCP Basic — $199 | |
|---|---|---|
| The problem it solves | Claude Code can do it, but re-asks permission, forgets the repo, or does it inconsistently. | Claude Code cannot do it at all, because nothing exposes it. |
| What you get | CLAUDE.md, a permission allowlist, hooks, 3–5 skills, up to 3 MCP servers wired from existing ones, and a runbook. | One MCP tool with a described schema, unhappy paths handled, a smoke prompt and an off switch. |
| Is it code? | Mostly configuration and plain text. | Yes. It is software, and it is tested like software. |
| Turnaround | About 48 hours. | A few days. |
| What it does not cover | Building anything new. | A second tool, or a Worker. That is a Build Packet, not a bigger MCP Basic. |
This is the line that causes the mix-up, so it is worth being exact. A Sprint wires MCP
servers that already exist — a filesystem server, a database server someone else published,
a vendor's own — into your .mcp.json with the right scopes and an off switch.
It writes no new server.
MCP Basic builds one: a tool that did not exist before, with a schema, error branches and tests. If you have been told “there is an MCP server for that” and there is, you want the Sprint. If you looked and there is not, you want MCP Basic.
You can see exactly what an MCP Basic hand-over looks like before you buy one: kit-one-tool-mcp is a working one-tool server, free and MIT, with 15 tests that run with no API key. It is the same shape, built against a made-up API instead of yours. Clone it and run the tests; that is the deliverable, unpaid.
npm test several times a weekStill stuck between them? Buy the Sprint. If it turns out you needed the tool built, reply to the receipt and say so, and you will be told the price from where you are before paying again. That is cheaper than guessing upward.
Sprint configures what Claude Code already has: memory, a permission allowlist, hooks, skills, and wiring for MCP servers that already exist. MCP Basic builds one MCP tool that did not exist, with a schema, handled error paths, a smoke prompt and an off switch. Configuration versus software.
Because those are different verbs. The Sprint wires servers that already exist into your config with the right scopes. MCP Basic writes a new server against an API that has none. If a server already exists for what you need, the Sprint covers it and the $199 would be wasted.
Yes. kit-one-tool-mcp on GitHub is a complete one-tool MCP server, free and MIT, with 15 tests that run with no API key and no network. It is the same shape as a paid build, made against a placeholder API instead of yours.
Reply to your receipt and say so. You will be told what the next rung costs from where you are before you pay anything again. Starting at the cheaper rung is usually the right move because the wrong guess upward costs more than the wrong guess downward.
No. MCP Basic is one tool. The Build Packet at $399 is an MCP server up to five tools, or a Cloudflare Worker up to three routes, with tests and a README. If you need more than one tool, do not buy MCP Basic twice.