What I use

Last updated: April 17, 2026

The current stack I reach for when building AI-native automation, agentic systems, MCP servers, and high-performance web applications. This page is rebuilt on every deploy, so it stays in sync with what I'm actually shipping — not an aspirational list.

AI & LLM Tooling

Claude (Anthropic) — daily driver for complex reasoning, code review, and long-context tasks. Claude Opus 4.7 with 1M-context window for whole-repo refactors.
Claude Code — CLI agent for actual codebase work. Where most of my shipped automation gets written.
Model Context Protocol (MCP) — the standard I bet on. I've shipped three open-source MCP servers (Ploomes, WhatsApp, ManyMCPs) and rely on community MCPs daily (Supabase, Figma, Gmail, Calendar, Playwright).
OpenAI API — for GPT-4.x when a specific task benefits from their tool use patterns.
Cursor / Zed — backup editors when I need a different keyboard flow.

Languages & Runtimes

TypeScript — default language for every shipping project. Strict mode, zod for runtime validation.
Node.js (v20 LTS) — MCP servers, CLI tools, serverless functions.
Python — data pipelines, ML experiments, scripting.
Bash / Zsh — daily automation glue.

Frontend

Astro.js v5+ — static-first with islands. Default choice for corporate sites, portfolios, and content-heavy pages. This site itself runs on Astro 6.
Next.js — when an app needs server actions, ISR, or edge-side rendering beyond static. Frontend shell for Supabase-backed dashboards.
Tailwind CSS v4 — utility-first styling, no framework overhead.
shadcn/ui — when Next.js; borrowed into Astro when needed.
Lucide Icons — consistent icon system.

Backend & Data

Supabase — PostgreSQL, auth, storage, realtime, edge functions. My default backend for anything that needs a database.
PostgreSQL — raw when I need performance tuning, pgvector for RAG.
Vercel — hosting, edge functions, Speed Insights, Analytics, Cron. Also where this site lives.
Trigger.dev v3 — code-first background jobs and long-running tasks replacing n8n for production workflows.

AI Infrastructure

Retrieval-Augmented Generation (RAG) — pgvector + Supabase for most cases; Pinecone when scale justifies it.
Embeddings: OpenAI text-embedding-3-large for quality, voyage-3 for cost-sensitive use cases.
Agent frameworks: Claude Agent SDK, MCP servers composed into larger pipelines, Vercel AI SDK for chat UIs.
Evaluation: lightweight eval harnesses in TypeScript — I avoid heavy frameworks until the problem justifies them.

CRM & Business Operations

Ploomes — the CRM I automate against most often (hence the unofficial MCP server).
n8n — still used for quick prototypes, then migrated to code-first when in production.
WhatsApp Business API — frequent integration target for LATAM clients.

Editor & Terminal

VS Code — primary editor with Claude Code, GitLens, ESLint, Prettier, Tailwind IntelliSense.
Ghostty — fast, modern terminal.
Zsh + Starship prompt — minimal, informative.
1Password CLI — secrets injection into dev environments without storing in .env files.

Design & UI

Figma — design source of truth for all client work.
Figma MCP — reads designs directly from Figma into Claude Code for implementation.

Dev Workflow

Git + GitHub — obvious.
pnpm — primary package manager.
Bun — for scripts that need speed.
GitHub Actions — CI; lightweight pipelines without over-engineering.
Playwright — end-to-end tests when the cost-benefit clears.