Stacks & Add-ons
Everything below comes from the template registry, so the exact catalog can grow without a CLI release. This is the current set.
Tech stacks
Each stack contributes an instructions section, a package.json fragment (where applicable), CI job fragments, a Dockerfile, and a specialist agent — retargeted to your selected tools. Stacks marked Starter scaffold a documented skeleton via the framework's official CLI; the rest ship runnable boilerplate.
Web
| Stack | Notes |
|---|---|
| React (Vite) | Runnable boilerplate |
| Next.js (App Router) | Runnable boilerplate |
| Vue (Vue 3 + Vite) | Runnable boilerplate |
| Svelte (Vite SPA) | Runnable boilerplate |
| SvelteKit | Meta-framework, adapter-node |
| Astro | Static-first site |
| Nuxt | Nuxt SSR |
| Angular | Starter |
Backend
| Stack | Notes |
|---|---|
| NestJS | Runnable boilerplate |
| Hono | Multi-runtime; build + runtime-smoke verified |
| Laravel (PHP) | Starter |
| Python FastAPI | Runnable boilerplate |
| Django (Python) | Minimal runnable (manage.py check) |
| Spring Boot (Java) | Starter |
| ASP.NET Core (.NET) | Starter |
Mobile & desktop
| Stack | Notes |
|---|---|
| React Native (Expo SDK 57) | Starter (full skeleton via the official CLI) |
| Flutter | Starter (full skeleton via the official CLI) |
| Tauri | Frontend-agnostic desktop shell |
Data apps
| Stack | Notes |
|---|---|
| Python Streamlit | Runnable boilerplate |
AI-native stacks
The on-brand core — scaffolds for building AI apps and agents, each defaulting to the latest Claude models.
| Stack | Notes |
|---|---|
| MCP server (TypeScript) | Model Context Protocol server; stdio-first, HTTP documented |
| MCP server (Python) | Same, Python SDK |
| Vercel AI SDK | Next.js chat UI, AI SDK v5 + Claude |
| LangGraph | Node agent, LangChain 1.0 createAgent + Claude |
| PydanticAI | Type-safe Python agents + Claude |
LLM provider, always
Every project's base instructions include an LLM-provider section (Anthropic / OpenAI SDK starter, default to the latest Claude) and a Deploying section (Vercel / Cloudflare Workers / Netlify / Railway / Fly / Render) — no separate selection needed.
Databases
Postgres · MySQL · MongoDB · Valkey (BSD-3 Redis-compatible) · Neon (managed serverless Postgres) · Supabase (Postgres + Auth + Storage) — each adds an instructions section, and self-hostable ones add a docker-compose service.
Vector databases
For RAG and AI-native apps — pick alongside a stack:
pgvector (default; drops into a Postgres service) · Qdrant (REST + gRPC) · Chroma. Each contributes an instructions section and, where self-hosted, a compose service.
ORMs
Prisma (v7 — prisma.config.ts + @prisma/adapter-pg driver adapter) · Drizzle (SQL-first). The ORM's dependencies are merged into your package.json and it adds an instructions section.
Storage
SeaweedFS (Apache-2.0, S3-compatible — self-host default, adds a compose service) · Cloudflare R2 (managed, $0 egress) · AWS S3 — instructions section per selection.
MinIO removed
MinIO was dropped over license risk (AGPL / likely source-available relicense). SeaweedFS is the self-hosted default; Cloudflare R2 the managed option.
Auth
JWT · Clerk · Better Auth · Auth.js (NextAuth) — instructions section plus verbatim setup files where applicable.
Infrastructure as Code
OpenTofu (MPL-2.0) — a .tf scaffold plus an instructions section. (Terraform is intentionally not offered — it's under the source-available BSL.)
CI
GitHub Actions · GitLab CI — the pipeline is composed per stack: a base skeleton plus one job per selected stack (or one job per app in a monorepo).
Docker
Opt in to generate a docker-compose.yml composed from a base plus one service per stack / database / vector-db / storage selection. Stacks use build: . against their per-stack Dockerfile; in a monorepo, each app gets its own build-based service.
Docs
Optionally scaffold a self-contained docs site into docs/ — choose Docusaurus, Fumadocs (Next.js), or Starlight (Astro). Each is its own sub-project — its package.json is copied, not merged into your root.
MCP servers
Model Context Protocol servers are the external tools your agent can call. Pick from a curated, developer-focused set:
| Server | What it gives the agent | Runs via |
|---|---|---|
| Chrome DevTools | Drive a real Chrome — DOM, console, network, performance, screenshots | npx chrome-devtools-mcp (no key) |
| Playwright | Drive a browser (Chromium/Firefox/WebKit) for e2e checks | npx @playwright/mcp (no key) |
| Context7 | Up-to-date, version-specific library docs pulled into context | remote HTTP (optional key) |
| GitHub | Repos, issues, PRs, code search, Actions | remote HTTP (OAuth) |
| Filesystem | Structured file read/search/edit scoped to the project | npx …/server-filesystem (no key) |
| Git | Local Git ops (status, diff, log, blame, commit) | uvx mcp-server-git (needs uv) |
| Fetch | Fetch a URL → clean Markdown | uvx mcp-server-fetch (needs uv) |
| Sequential Thinking | Structured, revisable step-by-step reasoning | npx …/server-sequential-thinking (no key) |
| Memory | Persistent knowledge-graph memory | npx …/server-memory (no key) |
Each selection is written to every selected tool's own MCP config, merged in so servers you already have are never overwritten:
- Claude Code →
.mcp.json(mcpServers) - OpenCode → the
mcpblock ofopencode.json - Codex →
.codex/config.toml([mcp_servers.*]) - Cline → not written (Cline's MCP settings live in IDE storage, not a project file) — add these from Cline's MCP panel.
A short section per server is also appended to your instructions file. Servers that need a key or runtime (Context7, GitHub, Git, Fetch) say so in that section; the configs are written so the zero-config ones run as-is.
The core set
Installed on every project regardless of selections (and hidden from the pickers):
- Skills —
engineering-standards,knowledge-base,using-create-ai-native-project - Agents —
code-reviewer,security-reviewer
These are the always-on quality and security guardrails.
Skills & agents (batch 1)
Beyond the core set, you can opt into a batch of reusable skills and specialist subagents:
- Skills —
test-generation,debugging,pr-description,refactoring,performance-profiling,dependency-upgrade - Agents —
test-engineer,api-designer,database-optimizer,performance-engineer,devops-engineer,docs-writer,accessibility-auditor
Each stack also auto-installs its own specialist agent (e.g. a react or django agent) when you select it.