CrabStack / Best for AI agents

The best SaaS boilerplates for AI coding agents (2026)

Disclosure: CrabStack publishes this page and ranks itself first. So you can judge it anyway, the criteria come before the ranking, every claim about the other products was checked against their own sites on October 28, 2026 and links to its source, and each entry says who should pick it. If this page and a vendor's site disagree, the vendor's site is right.

What makes a boilerplate good for an AI agent

"AI-ready" is on every boilerplate's homepage now. These are the properties an agent can actually use, most important first:

  1. Tests the agent can run. An agent that can check its own change catches its own mistakes, and tests on the code that handles money matter most.
  2. One way to do each thing. Agents copy what they see, so a codebase with three ways to write an endpoint soon has a fourth.
  3. A contract the agent can read. An API description it can check its work against, verified in CI so it cannot drift from the code.
  4. Instructions written for agents. AGENTS.md, CLAUDE.md and skill files tell the agent where things go. Every kit below except ShipFast documents them, so on their own they no longer decide much.
  5. Docs that explain why. When the docs give the reason for a design, an agent that extends it keeps the intent.

The ranking

1. CrabStack: $199 or $349 one-time, before tax (ours, so check it against the criteria)

A NestJS 12 backend, a Next.js 16 frontend and PostgreSQL. It ships 1,000+ tests, including the Stripe webhook cases that cost money (duplicates, events out of order, disputes and refunds), one folder shape for every new feature, an OpenAPI contract checked in CI, AGENTS.md, CLAUDE.md files and a skill for adding a feature, and public docs with step-by-step recipes. Its MCP server runs inside the deployed app; Makerkit's runs on your machine while you build, and supastarter and ShipFast advertise none. crabstack.dev runs on the template. Not included: file uploads, websockets, MFA, a third-party UI library, a database, schema or domain per tenant (workspaces share one database), languages beyond English (i18n is set up, English ships). If you need any of those on day one, CrabStack is the wrong pick.

2. Makerkit: $349 or $649 one-time, shown at $299 or $599 on October 28, 2026

The broadest kit: teams, roles, MFA, multiple languages, Shadcn UI and Figma files, on Supabase, Drizzle or Prisma, with Next.js, TanStack Start or React Router. For agents it has AGENTS.md and CLAUDE.md files, rules for Claude Code, Cursor, Codex and Gemini, skills such as /rls-review, Playwright and Vitest tests, and a development MCP server with 56 tools in its 2.0 release. Pick it when you want Supabase, MFA or several languages on day one.

3. supastarter: $349, $799 or $1,499 one-time

A TypeScript monorepo with a Hono API and a choice of Next.js, Nuxt or TanStack Start, plus Better Auth, Prisma or Drizzle, five payment providers, organizations, multiple languages, file storage and an AI chatbot. For agents it has AGENTS.md and 22 skills in the Next.js kit, including skills that write tests and verify changes. Pick it for framework choice, or usage and seat billing. It does not state how many tests ship with it.

4. ShipFast: $299 or $349 one-time, shown at $199 or $249 on October 28, 2026

A starter for getting a landing page with sign-in and payments live quickly: Next.js API routes, MongoDB or Supabase, Stripe or Lemon Squeezy, and UI components. Its site and docs do not mention agent files, a test suite or an API contract, though its creator has published AI rules for Claude Code and Cursor. On the $199 Starter plan, lifetime updates are marked as not included, though its FAQ says buyers get lifetime updates. Pick it for a fast launch and its maker community.

Sources (read October 28, 2026)

  • CrabStack: docs
  • Makerkit: site, pricing, testing, MCP server 2.0
  • supastarter: site, agent skills, API overview
  • ShipFast: site, docs, Marc Lou on X
  • Claude Code build cost: AI Builder Club, April 12, 2026

FAQ

Should I have my agent build the foundation from scratch instead? For a prototype, yes. Generating the code is cheap: one published Claude Code build of sign-in, Stripe billing, a dashboard and deployment took about 8 hours and about $12 of API usage (AI Builder Club, April 12, 2026). The work is in knowing the failure modes, such as webhooks that arrive twice or out of order, a chargeback, or a worker that dies halfway, and in proving the code handles them. That is why tests come first in the criteria.

Why trust a ranking published by one of the products? Don't take it on trust. Check whether the criteria make sense, whether the claims about the other products match their sites today, and whether the page says what its own product lacks. This page is written to pass that check, and if another vendor's page passes it too, weigh both equally.