BriefKit MCP Server
mithun4elp/briefkit-mcp-server
Specs de SaaS em qualidade engenharia, 70-85% menos tokens
Ver no GitHubServidor MCP que gera especificações de SaaS prontas para produção (design system, schema PostgreSQL, políticas RLS) diretamente dentro de Claude e outros assistentes IA. Reduz consumo de tokens em 70-85% ao fornecer ferramentas estruturadas para geração de specs técnicas, eliminando raciocínio redundante do modelo.
Instalação
npx convez-ai mithun4elp/briefkit-mcp-server1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Gerar schemas PostgreSQL completos com tabelas, FKs e segurança baseline
- Criar políticas RLS (Row Level Security) para múltiplos papéis em Supabase
- Produzir design systems estruturados (cores, tipografia, componentes, dark mode)
- Acelerar prototipagem de SaaS reduzindo iterações e consumo de tokens
Veredito

Ferramenta prática para quem constrói SaaS com IA — especializada e eficiente. A redução real de tokens (11.200 → 1.800 em um caso CRM) é comprovada, mas a utilidade está fortemente atrelada a workflows que precisam de specs técnicas repetitivas.
README
A tradução para o português deste README ainda está na fila.
BriefKit MCP Server
Engineer-grade SaaS specification generator for AI build tools.
What it does
This MCP server lets AI assistants (Claude, ChatGPT, etc.) generate structured SaaS specifications on demand. When a user asks about building a SaaS product, the AI can call these tools to produce production-ready specs.
Token efficiency
BriefKit MCP reduces token consumption by 70-85% for SaaS specification tasks on Claude Code, Cursor, and other AI assistants.
Without MCP: Claude reasons through schema, RLS, and design decisions from scratch — typically 8,000-15,000 output tokens per response.
With MCP: Claude calls structured tools that return pre-built specs in 500-1,500 tokens.
Real measurements (CRM for real estate agents):
- Without MCP: ~11,200 tokens, 4 turns to converge
- With MCP: ~1,800 tokens, 1 turn
If you're hitting Claude Code usage limits faster than expected on spec-heavy work, this MCP keeps spec-generation tasks under 2,000 tokens per call.
Tools
| Tool | What it generates |
|---|---|
briefkit_generate_design_system | Complete DESIGN.md — colors, fonts, spacing, components, dark mode |
briefkit_generate_database_schema | PostgreSQL/Supabase schema — tables, columns, types, FKs, indexes, security baseline |
briefkit_generate_rls_policies | Row Level Security policies — per table, per role, with server-only restrictions |
briefkit_get_info | Information about BriefKit and the full 14-file brief pack |
Setup
Install
git clone https://github.com/briefkit/briefkit-mcp-server.git
cd briefkit-mcp-server
npm install
npm run build
Use with Claude Desktop
Add to your claude_desktop_config.json:
{
"mcpServers": {
"briefkit": {
"command": "node",
"args": ["/path/to/briefkit-mcp-server/dist/index.js"]
}
}
}
Use with Claude Code
claude mcp add briefkit node /path/to/briefkit-mcp-server/dist/index.js
Examples
User: "I'm building a CRM for real estate agents. Help me set up the database."
AI calls: briefkit_generate_database_schema with:
- product_name: "RealEstCRM"
- saas_type: "b2b"
- custom_tables: ["properties", "showings", "offers", "commissions"]
Result: Complete SQL schema with 10 tables, typed columns, foreign keys, indexes, and security baseline.
User: "Generate RLS policies for my SaaS. I have profiles, orders, and payments tables with Owner and Member roles."
AI calls: briefkit_generate_rls_policies with:
- tables: ["profiles", "orders", "payments"]
- roles: ["Owner", "Member"]
Result: Complete RLS SQL — payments are server-write-only, profiles role is immutable, owners see own data only.
Full Brief Pack
These tools generate individual specification components. For the complete 14-file engineer-grade brief pack (design system + schema + RLS + auth + billing + 47 test cases), visit:
briefkit.online — $9 per version
Free Tools
License
MIT