convez.ai
MCP ServerDevOps & InfraestruturaIntermediário
illinigirl

Claude Code Toolkit

illinigirl/claude-code-toolkit

Gere, valide e publique MCP servers prontos para produção com um comando

Ver no GitHub
InicialPluginScore · de 0 a 100

Suite de skills e MCP servers para Claude Code que automatiza a geração, validação e publicação de servidores MCP. Inclui ferramentas para scaffolding de projetos prontos para produção, verificação de saúde, auditoria de segurança e handoff de sessões.

Instalação

$npx convez-ai illinigirl/claude-code-toolkit

1 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

Estrelas
0
Forks
0
Acompanhando
0
Saúde
42%
Último commit
há 4 meses
Versão
—
Issues abertas
0
Contribuidores
1

Casos de uso

  • Gerar skeleton de MCP server com testes e CI já configurados
  • Verificar saúde e integridade de servidores MCP existentes
  • Auditar repos antes de publicação pública (secrets, PII, LICENSE)
  • Documentar design e handoff de sessões entre contextos

Veredito

Toolkit sólido que reduz atrito no ciclo de vida de MCP servers, com geração de projetos production-ready e validação integrada. Essencial para quem constrói servidores MCP com frequência ou quer garantir qualidade em publicações.

README

A tradução para o português deste README ainda está na fila.

claude-code-toolkit

A small toolkit of Claude Code skills, distributed as an installable plugin marketplace. Skills and a failure-mode hook today; room to grow into agents and MCP servers.

Skills

SkillCommandWhat it does
scaffold-mcp/scaffold-mcpGenerate a new Python MCP server — a runnable, tested, ruff-clean skeleton in the pure-core + thin-adapters shape (FastMCP, dual stdio/HTTP transport, CLI adapter, one I/O module, bundled seed, tests, CI, CLAUDE.md, pyproject).
verify-mcp/verify-mcpHealth-check an MCP server: venv + ruff + pytest, import the server to prove its tools register, inventory the @mcp.tool() surface (count + missing docstrings — the contract Claude reads), then report a GREEN/RED verdict with the specific cause on failure. Scaffold builds it; verify proves it.
public-ready/public-readyAudit a repo for public release — scan tracked files for secrets, PII, and interview/portfolio framing; verify tests + lint are green; check LICENSE / README / .gitignore / CI — then publish (gh repo create) on your approval.
handoff/handoffWrite a PICKUP-*.md session handoff (task, what's done, what's in flight, next steps, key paths, working norms) so a fresh session resumes cold. Stays local, never committed.
failure-scan/failure-scanReview the current diff against the failure-mode catalog — the judgment classes a hook can't catch (plausible-but-wrong AI output, biased per-segment stats, dispatch-order coupling) — reporting concrete risks with each class's verify questions and fix pattern.
design-note/design-noteWrite a structured <feature>-DESIGN.md before building — problem, the core reframe, the design with options + tradeoffs, edge cases, and open questions to decide at build time.
note/noteFrictionless note-to-self — dump a thought and it's auto-classified into a bucket (note/skill/mcp/todo) under ~/.claude/notes/, no category-picking. Bare /note lists everything; /note <bucket> lists one.

What /scaffold-mcp generates

It produces a complete, runnable project — not a snippet. One invocation gives a tested, ruff-clean MCP server skeleton that passes its own suite out of the box:

$ /scaffold-mcp   (→ scaffold.py --name trip-logger --domain trip)

trip-logger/
├── src/triplogger/
│   ├── server.py      FastMCP server — dual stdio / HTTP transport
│   ├── core.py        pure domain logic, no I/O — the part you replace
│   ├── store.py       JSON-file persistence (the one I/O module)
│   ├── cli.py         thin CLI adapter over the same core
│   ├── data/…seed.json   bundled seed, packaged so any install runs immediately
│   └── models.py · exports.py · __init__.py
├── tests/             conftest + core/server tests (green on generation)
├── .github/workflows/ CI (pytest)
├── CLAUDE.md          orientation: architecture + where to extend
├── setup.sh · preflight.sh   one-command venv setup + lint/test gate
└── pyproject.toml · requirements.txt · .gitignore · LICENSE · README.md

$ pytest -q      →  17 passed
$ ruff check .   →  All checks passed!

You swap in your own domain logic (the skill leaves TODO(domain) markers); the structure, both transports, tests, CI, and packaging are already done.

Failure-mode hook + /failure-scan

A small system for classes of bugs you've hit before — so you catch them by default, not by remembering. Two halves over one catalog (failure-modes/catalog.md):

  • A PostToolUse hook (auto-registered on install) flags the grep-able classes the moment you edit a file — e.g. a DynamoDB Scan / FilterExpression with no pagination, which works until a table outgrows one page and then silently returns partial results. Non-blocking, fail-safe (never disrupts an edit), and intentionally high-signal — a noisy hook gets ignored.
  • /failure-scan reviews the current diff against the judgment classes a regex can't catch — plausible-but-wrong AI output, biased per-segment stats, implicit dispatch-order coupling — using each class's verify questions.

Add a class once to catalog.md, tagged hook (gets a grep rule in rules.json) or judgment (reviewed by the skill). failure-modes/test_hook.py keeps the hook honest.

Install (as a plugin)

/plugin marketplace add illinigirl/claude-code-toolkit
/plugin install claude-code-toolkit@claude-code-toolkit

Skills then invoke under the plugin namespace, e.g. /claude-code-toolkit:scaffold-mcp. Run /plugin marketplace update claude-code-toolkit to pull updates.

Or use a single skill directly

Skills are also plain ~/.claude/skills/ directories — symlink or copy one in:

ln -s "$PWD/skills/scaffold-mcp" ~/.claude/skills/scaffold-mcp

Then invoke it by name, e.g. /scaffold-mcp. A symlink during development means edits take effect live.

Layout

claude-code-toolkit/
  .claude-plugin/
    marketplace.json    catalogs the plugin (so it's /plugin install-able)
    plugin.json         the plugin manifest
  skills/
    scaffold-mcp/       SKILL.md · reference.md · scaffold.py · templates/
    verify-mcp/         SKILL.md — health-check an MCP server (GREEN/RED verdict)
    public-ready/       SKILL.md · audit.sh
    handoff/            SKILL.md · template.md
    failure-scan/       SKILL.md — review a diff against the catalog
    design-note/        SKILL.md — write a <feature>-DESIGN.md
    note/               SKILL.md — frictionless note-to-self capture
  hooks/
    hooks.json          registers the failure-mode PostToolUse hook
  failure-modes/
    catalog.md          the bug-class catalog (the shared brain)
    rules.json          grep-able rules the hook runs
    hook.py · test_hook.py
  .github/workflows/
    test.yml            CI — scaffolds + verifies a generated project end-to-end,
                        tests the worked example (and drives its demo), runs the
                        hook regression suite, validates manifests + frontmatter
  README.md · LICENSE

Validate the manifests with claude plugin validate ..

License

MIT.

Ver o original no GitHub

Bom para

Relacionados