Trabuco MCP server

████████╗██████╗ █████╗ ██████╗ ██╗ ██╗ ██████╗ ██████╗ ╚══██╔══╝██╔══██╗██╔══██╗██╔══██╗██║ ██║██╔════╝██╔═══██╗ ██║ ██████╔╝███████║██████╔╝██║ ██║██║ ██║ ██║ ██║ ██╔══██╗██╔══██║██╔══██╗██║ ██║██║ ██║ ██║ ██║ ██║ ██║██║ ██║██████╔╝╚██████╔╝╚██████╗╚██████╔╝ ╚═╝ ╚═╝ ╚═╝╚═╝ ╚═╝╚═════╝ ╚═════╝ ╚═════╝ ╚═════╝

Generate production-ready Java projects — with the AI context your coding agents need built in.

Starting a new Java project is slow — even with a coding agent helping. You set up Maven modules, wire databases and tests, copy boilerplate from that one project that "mostly works," and then spend the first hour explaining every architectural convention to Claude Code, Cursor, Copilot, or Codex from scratch. Hours pass before you write a single line of business logic. Trabuco exists because your time — and your agent's context window — is better spent building features, not fighting configuration or re-teaching conventions.

Trabuco is a command-line tool — and a Claude Code plugin — that generates both halves of a modern Java codebase: a complete, production-ready multi-module Maven project and the AI context that teaches coding agents how to work in it. Run trabuco init (or inside Claude Code, type /trabuco:new-project and describe what you need in plain English), answer a few prompts (or pass flags for automation), and you get a fully wired Spring Boot codebase alongside task-specific prompts, quality specifications, per-agent rule files, and workflow hooks already configured for Claude Code, Codex, Cursor, and GitHub Copilot. No templates to download, no manual setup, and no session spent bootstrapping your agent's understanding of the project.

The generated code is production-grade by default. Spring Boot with Spring Data JDBC (no JPA surprises), Flyway migrations, Testcontainers for real integration tests, Resilience4j circuit breakers, Google Java Format enforced by Spotless, ArchUnit rules that fail the build on layer violations, correlation-ID tracing, Prometheus metrics, OpenAPI + Swagger UI, and a global exception handler with sanitized responses. PostgreSQL, MySQL, MongoDB, or Redis — all configured with Docker Compose. JobRunr for background jobs; Kafka, RabbitMQ, AWS SQS, or GCP Pub/Sub for event-driven processing. The modular layout — Model, SQLDatastore / NoSQLDatastore, Shared, API, Worker, EventConsumer — has clean compile-time boundaries so API physically cannot import Worker code. Every opinion is deliberate: keyset pagination, no foreign-key constraints, Immutables at module boundaries, constructor injection only, bulk-bounded writes.

Alongside the code, Trabuco lays down an AI collaboration layer that the major coding agents load natively. The .ai/prompts/ directory ships task-specific guides (add-entity, add-endpoint, add-service, add-event, add-job, add-tool) plus JAVACODEQUALITY.md — an authoritative specification covering architecture boundaries, exception handling, datastore performance (bulk I/O, keyset drain loops, denormalization), and testing standards. Per-agent rule files — CLAUDE.md for Claude Code, AGENTS.md for Codex, .cursor/rules/java.mdc for Cursor, .github/instructions/java.instructions.md for Copilot — wire those conventions into each tool's native discovery. Claude also gets .claude/skills/ for commit, PR, and review workflows; Codex and Cursor get hooks; Copilot gets setup steps. Every architectural convention lives in two places: enforced by the generated code and explained to the agents that will extend it.

Shortened. The full README is on GitHub.