Two jobs an AI SaaS starter kit has to do
An agent can scaffold a chat UI and an OpenAI call in minutes. That part of an AI SaaS was never the hard part. The hard parts are the ones that break quietly: a query that returns another team's documents, a webhook that misses a plan change, a user who burns through tokens nobody bills for.
So the kit does two things. It gives your coding agent the context and tools to follow the codebase's conventions instead of inventing new ones. And it gives your product the multi-tenant, billing-aware baseline that AI features sit on.
A SaaS starter kit with AI agent tooling built in
Agents read your schema instead of reconstructing it from fifteen migration files, and follow the kit's patterns for Server Actions, RLS, and permissions.
Agent rules in the repository
A root AGENTS.md plus package-level files for the app, database, UI, and feature packages. Agents working in the Supabase folder read the migration and RLS rules; agents in the UI package read the component rules. CLAUDE.md, .gemini and .agents configs point the other tools at the same files.
MakerKit MCP server
An MCP server inside the monorepo that lets the agent inspect your current database schema, generate and apply migrations, read and validate environment variables, start dev services, read emails from the local mailbox, and manage translations. MCP 2.0 (February 2026) shipped 56 kit tools.
CLI MCP server
A second MCP server from the MakerKit CLI that installs plugins, pulls upstream kit updates, and helps resolve merge conflicts, so staying current is a task you can hand to the agent.
Works with the tools you use
Configured for Claude Code, Cursor, Windsurf, Codex, Gemini, and Google Antigravity. The MCP server works with any MCP-compatible agent.
Setup is in the docs for AI agent rules and the MakerKit MCP server. The MCP 2.0 release notes list every tool.
An AI SaaS template is only as good as the SaaS under it
AI products cost money per request and handle customer data. Both problems are solved below the AI layer, and this is the layer MakerKit ships finished.
Team workspaces with RLS
Personal and team accounts, invitations, and roles. In the Supabase kits, account data is protected by Postgres Row Level Security, and new tables follow the same account_id and policy pattern, so which team sees which prompts, files, and outputs is decided by the database, not by each query an agent writes.
Usage-based billing
Metered usage for Stripe and Lemon Squeezy is built in: report consumption and the provider invoices it. Per-seat and flat subscriptions work alongside it.
Credits and token limits
A documented recipe: a credits table per account, a database function that consumes credits, and a webhook that recharges them on renewal. Credits are not built in, so you add only the tables your pricing needs.
Remote MCP for your customers
If your customers want to connect Claude or Cursor to your product, our tutorial builds a remote MCP server with OAuth 2.1 on Supabase Auth, scoped per workspace and enforced by RLS.
Read the guides on metered usage billing, credit-based billing for AI SaaS, and building an MCP server with OAuth.
Our customers love Makerkit ❤️
Rob, founder at Provider.app
Watch the full interview with Rob.Sebastian, founder at Watchthis
Watch the full interview with Sebastian.Michael, founder at ChatFlow
Watch the full interview with Michael.One license, agent tooling included
One-time payment, lifetime updates, unlimited projects. Every stack ships the same MCP server, agent rules, and metered billing.
- Lifetime access to your stack of choice
- Build unlimited applications
- Continuous updates
- Access to the Discord community
- The best support in the SaaS Starter kits market, period.
- Community and Chat support
- Community-based feature requests and Request for Comments. You have a say in what Makerkit should build next.
- One-time payment, no subscriptions
- Includes access to our free Figma UI kit
Choose your Stack
Select the technology stack that best fits your project
AI SaaS starter kit FAQ
What is an AI SaaS starter kit?
A codebase that gives you the non-AI parts of an AI product (auth, team workspaces, billing, admin) so you can spend your time on the model integration and the product. MakerKit also ships an MCP server and agent rules so AI coding agents build on the kit correctly.
Is MakerKit an AI SaaS template?
It is the part of an AI SaaS template that decides whether you can charge for it: auth, team workspaces isolated per account, subscriptions and metered billing, and an admin panel, plus the MCP server and agent rules that keep your coding agent on those patterns. It does not ship a finished AI app to reskin; your agent builds the AI features on top.
How do I charge for AI usage?
Metered usage billing is built in: you report usage to Stripe or Lemon Squeezy and the provider invoices at the end of the period. Credit or token based billing is not built in, but the docs include a recipe that adds a plans table, a credits table, database functions to consume credits, and a webhook that recharges credits on renewal.
Which AI coding agents does MakerKit support?
The kit ships AGENTS.md and CLAUDE.md files plus configuration for Claude Code, Cursor, Windsurf, Codex, Gemini, and Google Antigravity. The MakerKit MCP server works with any MCP-compatible agent.
Is the MCP server included in every license?
Yes. The MCP server and agent rules are included in both the Pro and Teams licenses, on every stack.
What the kit does not include
- No chatbot or RAG pipeline: no LLM provider integration, chat UI, embeddings, or vector store. Add the SDK you prefer, such as the Vercel AI SDK, OpenAI, or Anthropic, inside the kit structure; the kit stays neutral on model and provider.
- No built-in credit system. Use metered billing, or follow the credits recipe.
- No ready-made AI app templates. The chat with PDF and blog writer demo codebases from 2023 and 2024 were built on earlier kit versions and are not part of the current kits.
Your agent can write the AI features. MakerKit is the part it shouldn't improvise: auth, team isolation, and billing that already work in production. See how that looks in practice in building a SaaS with Claude Code.