Agentloom
farnoodma/agentloom
Defina seus agentes uma vez, use-os em qualquer ferramenta de IA
Ver no GitHubMCP server que unifica a configuração de agentes, comandos, skills e servidores MCP em um único diretório `.agents/`, sincronizando automaticamente para múltiplos provedores (Cursor, Claude, Copilot, Codex, OpenCode, Gemini, Pi). Elimina duplicação de configurações e mantém uma única fonte de verdade versionável.
Instalação
npx convez-ai farnoodma/agentloom1 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
- Sincronizar prompts e configurações entre Cursor, Claude, Copilot e outras ferramentas de IA
- Compartilhar agentes e skills com equipes via GitHub
- Manter versionamento centralizado de agentes e MCP servers em repositório git
Veredito

Solução pragmática para fragmentação de ferramentas de IA, oferecendo sync bidirecional e portabilidade real sem lock-in. Bem-estruturado com suporte robusto a múltiplos provedores, mas requer curva de aprendizado do formato unificado.
README
A tradução para o português deste README ainda está na fila.
agentloom
Write your agents once. Use them everywhere.
If you're juggling Cursor, Claude, Copilot, Codex, Gemini, OpenCode, and Pi — you know the pain. Each tool has its own config format, its own folder structure, its own way of defining agents, commands, rules, and MCP servers. You end up copy-pasting prompts, maintaining multiple versions of the same setup, and losing track of what's where.
Agentloom fixes that. You define your agents, commands, rules, skills, and MCP servers once in a single .agents/ directory, and agentloom syncs them to every tool you use.
npx agentloom init
That's it. Agentloom detects your existing provider configs, migrates them into a unified canonical format, and syncs everything back out. Your agents now work across all your tools.
What you get
- One source of truth — a
.agents/directory with your agents, commands, rules, skills, and MCP configs in plain markdown and JSON. Version-controlled, diffable, reviewable. - Instant sync — run
agentloom syncand your definitions flow to Cursor, Claude, Copilot, Codex, OpenCode, Gemini, and Pi in their native formats. - Import from anywhere —
agentloom add user/repopulls agents and skills from GitHub repos. Share your best setups with your team or the community. - No lock-in — switch tools tomorrow and your agents come with you.
Quick start
# initialize and sync to your tools
npx agentloom init
# import agents from a GitHub repo (syncs automatically)
npx agentloom add farnoodma/agents
# re-sync after manual edits to .agents/
npx agentloom sync
Documentation
For full CLI usage, commands, schemas, and configuration details, see the CLI documentation.
Supported providers
| Provider | Agents | Commands | Rules | Skills | MCP |
|---|---|---|---|---|---|
| Cursor | + | + | + | + | + |
| Claude | + | + | + | + | + |
| Copilot | + | + | + | + | + |
| Codex | + | + | + | + | + |
| OpenCode | + | + | + | + | + |
| Gemini | + | + | + | + | + |
| Pi | + | + | + | + | + |
Directory
Browse and discover community agents, skills, commands, rules, and MCP configs at agentloom.sh.
Monorepo layout
packages/
cli/ # npm package: agentloom
apps/
web/ # Next.js directory + telemetry ingest API
Telemetry
Successful GitHub-based agentloom add imports can send anonymous install telemetry to the Agentloom directory API.
- tracked: agents, skills, commands, rules, and MCP servers from GitHub sources
- not tracked: local-path adds
- opt out:
AGENTLOOM_DISABLE_TELEMETRY=1 - override endpoint:
AGENTLOOM_TELEMETRY_ENDPOINT=https://...
Development
pnpm install
pnpm check
pnpm test
pnpm build
Run CLI from source:
pnpm --filter agentloom dev -- --help
Run web app:
pnpm --filter @agentloom/web dev
Release and deploy
- preview web deploys: on
pushtomain(.github/workflows/preview-deploy.yml) - production web deploy + npm publish: on GitHub
release.published(.github/workflows/release.yml) - stable tags only:
vX.Y.Z - release tag must match
packages/cli/package.jsonversion - Vercel auth in workflows: OIDC (no long-lived
VERCEL_TOKENGitHub secret required)
Required Vercel env vars:
DATABASE_URLTELEMETRY_HASH_SALT
License
MIT