saasprokit
Last verified: 25 August 2026. Every price and claim below is checked against the kit's own public site and docs.
Every SaaS boilerplate now calls itself "AI-optimized." Agent rules, MCP servers, LLM-friendly docs, the works. Here is the buyer's problem: most of these kits are paid and closed source, so you cannot inspect the repo before you pay, and the AI claims are exactly the kind of marketing that is cheap to write and expensive to discover was hollow.
This post does two things. First, it reviews what each major kit publicly publishes as evidence for its AI claims, kept strictly to what you can check yourself from the outside. Second, it gives you the checklist I would use to verify any kit's AI story before spending money, including on mine.
Full disclosure: I make one of these kits (SaaSProKit, $200). I have marked every place where that bias could matter, and nothing below about a competitor goes beyond their own public materials.
The claims, side by side
| Kit | Price | Agent rule files | MCP server | llms.txt | Multi-tenancy | Best for |
|---|---|---|---|---|---|---|
| ShipFast | $199-299 | Site says it gives AI editors context; no published rules I could find | Not claimed | Not claimed | No | Solo B2C, biggest community |
| MakerKit | $349-649 | Yes: published rules for Claude Code, Cursor, Codex, Gemini | Yes, documented | Yes | Yes | B2B teams, longest track record |
| supastarter | $299-1,499 | Yes: AGENTS.md, Agent Skills, Cursor/Windsurf/Claude Code | Not claimed | Yes | Yes | Funded teams, maximum breadth |
| SaaSProKit | $200 | Yes: AGENTS.md, 11 agent skills, and agent rules in the repo | No, and I will not pretend otherwise | Yes | Yes | Indie multi-tenant, agent-first |
Two things jump out of that table. Only three of the four kits ship multi-tenancy at all, and the AI-claim evidence ranges from "documented tooling you can read today" (MakerKit, supastarter) to "a sentence on a landing page" (ShipFast). Price does not predict either.
What "AI-native" should actually mean
Marketing has stretched the term to uselessness. When I say a codebase is AI-native, I mean four checkable things:
1. Agent rule files that exist and are current. An AGENTS.md, CLAUDE.md, or cursor rules file that describes the real architecture and conventions, not a README copy from three versions ago. This is the difference between an agent that extends your patterns and an agent that invents its own.
2. An MCP server, if claimed. Tooling the agent can query for accurate, versioned context about the kit. Only MakerKit publicly documents one today, and credit to them for it.
3. LLM-readable docs. An llms.txt or equivalent, so agents pull accurate context about the kit instead of scraping a marketing site. This one you can check from outside for any kit: try the domain followed by /llms.txt right now, before buying anything.
4. Codebase consistency. The quiet one that matters most, and the one no landing page can prove. Agents pattern-match. A repo where every feature follows the same shape lets an agent extend it correctly. A repo with three eras of conventions produces confident garbage. Rule files help an agent read the map; consistency is whether the map matches the territory.
Kit by kit, on public evidence
ShipFast ($199-299)
The site says the boilerplate gives your AI code editor real context to build full features. I found no published agent rules, no MCP server claim, and no llms.txt. That does not mean the repo is agent-hostile, and ShipFast's genuinely small codebase probably fits in an agent's context window better than most. It means the AI claim is thin on public evidence, so if agent workflow is why you are buying, ask in their community for specifics before paying. Where ShipFast is strong has never been AI anyway: it is the cheapest known brand with the biggest community, for single-user B2C products. My full comparison: SaaSProKit vs ShipFast.
MakerKit ($349-649)
The most substantiated AI story in the market, and I say that as a direct competitor. Published rules for Claude Code, Cursor, Codex, and Gemini, a documented MCP server, a public llms.txt, plus 400+ pages of docs an agent (or you) can actually use. If your budget covers $349 and you want the deepest paper trail behind the AI claims, this is the honest pick. What the public materials cannot tell you is how their seven web kits (Next.js, TanStack Start, and React Router across Supabase, Drizzle, and Prisma) affect an agent that has to figure out which variant it is standing in, which is the one structural question I would ask in their Discord before buying. My full comparison: SaaSProKit vs MakerKit.
supastarter ($299-1,499)
Ships AGENTS.md and Agent Skills, documents features in LLM-friendly format (including a public llms.txt), and supports Cursor, Windsurf, and Claude Code. A serious, real commitment. The open question is the same one that runs through everything about supastarter: it is a multi-package Turborepo across up to three frameworks, which is more surface for an agent to load correctly before its first edit. More context is not automatically worse, but it is the thing I would test in their demo before buying. My full comparison: SaaSProKit vs supastarter.
SaaSProKit ($200)
Mine, so maximum suspicion applies, and here is the claim stated plainly so you can hold me to it: one stack per kit (Next.js, Prisma, Postgres today; a TanStack Start kit ships September 2026), multi-tenancy with organizations, invitations, and RBAC, an admin panel, 2FA, locale routing for six languages, Stripe billing, and AGENTS.md, 11 agent skills, and llms.txt. No MCP server yet, no background jobs, and a community that is new and small. The bet is that one stack variant, kept consistent, serves an agent better than tooling layered on a large multi-variant codebase. If that bet is wrong for your use case, MakerKit's MCP server or supastarter's Agent Skills are the right call, and I would rather you buy the right kit than mine.
How to verify any kit's AI claims before you pay
Five checks, all free:
- Visit the domain followed by /llms.txt. Exists and current, or 404.
- Search their docs for "Claude Code", "Cursor", "agent". Real setup pages, or one marketing sentence.
- Read the changelog. Do AI-related files get updated when features ship, or were they written once for launch week.
- Ask in their Discord: "what does your AGENTS.md cover?" A maintainer who wrote one answers in detail. A maintainer who did not changes the subject.
- If there is a free tier or demo repo (MakerKit Lite, open-source kits), clone it and point your agent at a small task before buying the paid version.
Ten minutes of this filters out most of the hollow claims in the market, mine included if I ever let those files rot.
Verdicts by buyer type
Solo dev, B2C, smallest budget: ShipFast at $199. No multi-tenancy, and the AI claim is lightly evidenced, so treat agent workflow as a bonus, not a reason to buy.
Indie dev, multi-tenant B2B, agent-first workflow: this is who I built SaaSProKit for. At $200 flat it sits between supastarter's Solo tier ($299 list, often discounted below mine) and MakerKit at $349, so the case is the codebase: one stack variant an agent can hold, without disambiguating frameworks. If you want an MCP server and the longest track record instead, spend the extra $149 on MakerKit and you will not regret it either.
Funded team, breadth and providers: supastarter. Three frameworks, five payment providers, i18n, background jobs. Priced for teams that will use it.
B2B with a security checklist and budget: MakerKit. The maturity, docs, and documented AI tooling are worth the premium at that profile.
FAQ
Which boilerplate works best with Claude Code? On public evidence, MakerKit has the most documented AI tooling (rules plus MCP plus llms.txt). SaaSProKit's bet is a smaller, single-variant codebase with AGENTS.md, 11 agent skills, and llms.txt. The honest answer is that "works best" depends on repo consistency, which no landing page proves, so run the five checks above on your shortlist.
Do I still need a boilerplate if AI writes the code? The 2026 question. Agents scaffold a prototype fast. What they do not hand you is a hardened auth, billing, and tenancy boundary that has survived other people's production traffic. The common failure mode is code that looks right and quietly is not: limits enforced in the UI instead of the backend, tenant checks missing on one route out of twenty. A boilerplate is not competing with your agent. It is the substrate that decides whether your agent's output is safe to ship.
Is ShipFast AI-optimized? Its public materials claim AI-editor context but publish no rules files, MCP, or llms.txt. Its small codebase is a genuine practical advantage for agents. If this matters to you, ask for specifics in their community before buying.
How often is this updated? Pricing and claims re-verified monthly, and whenever a kit ships a major AI-related change. Corrections: ping me on X @luannguyenbkit, fixed with credit.
I sell one of the kits on this page. Every competitor claim above is taken from their own public site and docs so you never have to take my word for anything.