convez.ai
MCP ServerDevOps & InfraestruturaAvançado
sshoecraft

ccenv - Ambiente Claude Code Empacotado

sshoecraft/ccenv

Harness integrado do Claude Code com extensibilidade via overlay

Ver no GitHub
InicialPluginScore · de 0 a 100

MCP server que agrupa cinco ferramentas de tooling do Claude Code (ccmemory, ccusage, ccteam, ccproject, ccloop) em um único repositório instalável. Oferece sistema de overlay para estender funcionalidades por usuário/host sem fazer fork, incluindo awareness de projeto, memória persistente com índice FTS5, uso de contexto em tempo real, e coordenação multi-instância.

Instalação

$npx convez-ai sshoecraft/ccenv

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 1 semana
Versão
—
Issues abertas
0
Contribuidores
1

Casos de uso

  • Gerenciar contexto e memória em sessões longas do Claude Code
  • Rastrear uso de contexto-window e rate limits em tempo real
  • Coordenar trabalho entre múltiplas instâncias Claude Code com NATS JetStream
  • Estender funcionalidades via overlay system sem modificar repositório base
  • Estruturar awareness de projeto com constituição → docs subsistema → mapa estrutural

Veredito

Solução sofisticada para operacionalizar Claude Code em ambientes complexos. Exige Python 3.11+, NATS JetStream (para ccteam) e entendimento técnico de MCP, mas oferece arquitetura extensível bem pensada via overlay system. Ideal para equipes que precisam coordenar múltiplas sessões e gerenciar estado compartilhado.

README

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

ccenv

Self-contained Claude Code env/harness. Bundles five Claude Code tooling projects into one installable repo, and ships an overlay system so a host or user can layer additional MCP servers and instructions on top without forking.

What's in here

ComponentWhat it doesMCP name
ccprojectThree-layer project awareness skill (constitution → subsystem docs → structural map) + global ~/.claude/CLAUDE.md snippet—
ccmemoryPersistent file-backed memory with FTS5 index, autoinstalled hooks, MCP serverccmemory
ccusageReal-time context-window + rate-limit usage as an MCP tool and statuslineccusage
ccloopRelay-loop wrapper that hands work between fresh Claude Code sessions as context fills—
ccteamMulti-instance coordination layer (filesystem replication via NATS JetStream, file-level locking, MCP tools + hooks)ccteam

Each component still lives in its own subdirectory and has its own README.md, pyproject.toml, and tests. Top-level install.sh delegates to each component's own installer where one exists, and falls back to pip3 install --user <path> (non-editable) for the rest.

Install

git clone https://github.com/sshoecraft/ccenv.git
cd ccenv
./install.sh

Per-component options:

./install.sh --skip ccteam        # skip a component (repeatable)
./install.sh --only ccmemory      # install only listed components (repeatable)
./install.sh --no-overlays        # skip the overlay scan
./install.sh -h                   # full help

Re-running is idempotent. Each sub-installer checks its own state; MCP registrations use claude mcp get <name> to detect prior installs; ~/.claude/CLAUDE.md overlay blocks are stripped and re-applied so stale content self-heals.

Overlay system

install.sh scans three locations for per-user / per-host extensions that should layer on top of the bundled components:

/usr/local/ccenv        — system-wide overlay
~/.config/ccenv         — per-user overlay
<this script's dir>     — bundled (only scanned for MCP subdirs, not CLAUDE.md)

In an overlay directory, the installer looks for:

  • CLAUDE.md — appended to ~/.claude/CLAUDE.md inside a # [CCENV OVERLAY: <path>] … # [/CCENV OVERLAY: <path>] marker block. Re-runs strip stale blocks and re-merge eligible ones. (System and user dirs only — never the bundled dir.)

  • <subdir>/pyproject.toml — pip-installed (--user) and registered as a user-scope MCP server. Default registration: name and command both default to the subdir name. Override with <subdir>/.ccenv-mcp.json:

    {"name": "myname", "command": "bin", "args": ["--flag"], "scope": "user"}
    

This lets you add custom MCP servers on a single laptop, or distribute them via /usr/local/ccenv on a shared box, without touching the bundled component tree.

MCP server naming

All MCP servers register at user scope under a cc<short> convention: ccmemory, ccusage, ccteam. Overlay MCP servers default to their subdir name unless overridden.

Requirements

  • Python 3.11+ (3.12 verified)
  • pip3
  • Claude Code CLI on PATH for MCP / hook registration steps (the installer falls back gracefully and prints warnings if it can't find claude)
  • nats-server with JetStream — only if you want ccteam's cross-instance coordination. Without it, ccteam boots in standalone mode and a SessionStart hook surfaces a one-line notice in any .ccteam/-bootstrapped project so you know sync is off.

License

MIT — see LICENSE.

Ver o original no GitHub

Bom para

Relacionados