Loom — Ecossistema Unificado para Agentes IA
tusharmewara/homebrew-loom
Camada unificada que sincroniza skills, sessões e MCP entre agentes IA
Ver no GitHubMCP server que funciona como camada compartilhada entre múltiplas ferramentas de IA (Claude Code, jcode, Goose, etc.), permitindo compartilhar skills, sessões e servidores MCP sem reconfiguração. Inclui motor de roteamento inteligente que escolhe o modelo certo por complexidade da tarefa e daemon de fundo para performance otimizada.
Instalação
npx convez-ai tusharmewara/homebrew-loom1 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
- Compartilhar MCP servers e skills entre múltiplas ferramentas de IA sem duplicação
- Exportar e retomar sessões de um agente em outro com histórico preservado
- Rotear tarefas automaticamente para o modelo certo baseado em complexidade
Veredito

Loom resolve um problema real de fragmentação no ecossistema de agentes IA modernos, oferecendo infraestrutura sólida para compartilhamento de contexto. A implementação em Shell é enxuta, mas o projeto é ambicioso — ideal para quem orquestra múltiplos agentes em produção.
README
A tradução para o português deste README ainda está na fila.
homebrew-loom
Homebrew formula for Loom — the shared ecosystem for AI coding tools.
The Problem
If you use multiple AI coding tools (Claude Code, jcode, Cursor, Goose, opencode, crush...), you've probably noticed they don't share anything. Skills you install in one are invisible to the rest. MCP servers need to be configured separately each time. Sessions are locked inside whichever tool created them.
Loom is the shared layer between all of them.
Quick Start
# Install
brew tap tusharmewara/loom
brew install loom
# Initialize the ecosystem (creates ~/.loom/)
loom init
# Optional-but-recommended: start the background daemon
brew services start loom
That's it. Run loom status to confirm everything is set up.
First Steps
Once Loom is installed, these are the first things to try:
# See what Loom knows about — tools, skills, MCP servers
loom scan
# Check the health of your setup
loom doctor
# List installable MCP servers from the registry
loom mcp list-known
# Install one (e.g., Playwright for browser automation)
loom mcp install playwright
# List registered skills
loom skills list
Everything you add (MCP servers, skills, tools) becomes available to every tool in your ecosystem — no per-tool configuration needed.
Real-World Scenarios
Share an MCP server across tools
# Install once
loom mcp install filesystem
# Use from any tool — Claude Code, jcode, Goose, etc. — all share it
# without needing separate configs
Export a session and continue elsewhere
# After a session in any tool, export it to Loom's shared storage
loom sessions list
loom sessions export <session-id> --output session.jsonl
# Resume it in a different tool with full history
loom sessions resume <session-id> --provider anthropic --model claude-sonnet-4
# With knowledge injection, Loom enriches the resumed session with
# relevant context from past sessions
loom sessions resume <session-id> --inject-knowledge
Route tasks to the right model automatically
# Loom scores complexity and picks the right tier
# simple fixes → Slm (fast/cheap)
# complex architecture → Flagship (slow/expensive)
loom routing evaluate "Write a Python script to parse this CSV"
# Override for a specific task when needed
loom routing override <task-id> --tier flagship --reason "needs deep reasoning"
Daemon
The daemon (loomd) is a background service that keeps the ecosystem hot —
shared MCP connections, session indexing, event bus, knowledge graph — so CLI
commands start instantly. It auto-shuts down after 5 minutes idle.
brew services start loom # Start
brew services stop loom # Stop
brew services restart loom # Restart
Without the daemon, loom still works — it just initializes subsystems on
demand, which is slightly slower.
What Loom Provides
| Feature | What it does |
|---|---|
| Skills | Write a skill once, use it from any tool. Skills are portable YAML files with structured prompts, auto-loaded by all tools. |
| Sessions | Export from one tool, resume in another. Full history + compression + compression for long sessions. |
| MCP Servers | Register once, share across every tool. Pooled connections with health checking. |
| Auth & Credentials | Encrypted credential vault with role-based ACLs and audit logging. |
| Hooks | Intercept tool actions (pre/post task, session events) with executable scripts. Signed for integrity. |
| Knowledge Graph | Entities and relationships extracted from session history. Powers search, recommendation, and context injection. |
| Model Routing | Automatic complexity scoring routes tasks to the right model tier (Slm / Mid / Flagship). |
| Workflows | Multi-phase task orchestration with phase transitions and manual overrides. |
Command Reference
Core
loom init Create ~/.loom/ tree
loom scan Display resource summary
loom status Show ecosystem state
loom doctor Deep diagnostics
loom exec <tools...> [-- args...] Execute tools with ecosystem
loom tui Launch terminal UI
MCP
loom mcp list List registered MCP servers
loom mcp show <id> Show MCP server config
loom mcp add <name> Add MCP server stub
loom mcp install <name> Install a known MCP server
loom mcp list-known List installable MCP servers
Skills
loom skills list List skills
loom skills show <name> Show skill details
loom skills install <path> Install skill from file
Sessions
loom sessions list List sessions
loom sessions show <id> Show session details
loom sessions export <id> Export session
loom sessions resume <id> Resume session
Routing
loom routing evaluate <prompt> Score prompt complexity
loom routing show Show routing config
Knowledge
loom knowledge build Build knowledge graph from sessions
loom knowledge query <entity> Query subgraph around entity
loom knowledge search <query> Search entities
Daemon
loom daemon start Start loomd
loom daemon stop Stop loomd
loom daemon status Show daemon status
Formula
| File | Purpose |
|---|---|
Formula/loom.rb | Homebrew formula with bottles for arm64_tahoe, tahoe, x86_64_linux |
scripts/update_formula_checksums.sh | Fetch bottles and update checksums after a release |
.github/workflows/bottles.yml | CI: build and publish bottles on release |
Filesystem
~/.loom/
├── config.toml # Main config
├── config/ # Layer overlays (routing, mcp, hooks, acp)
├── skills/ # Shared skills database
├── mcp/ # MCP server configs
├── sessions/ # Session exports
├── auth/ # Encrypted vault + audit logs
├── hooks/ # Hook TOML configs and scripts
├── knowledge/ # Knowledge graph (SQLite + FTS5)
└── tools/ # Tool registration files
License
MIT