convez.ai
MCP ServerProdutividadeIntermediário
kumbukum

Kumbukum

kumbukum/kumbukum

Camada de memória compartilhada e pesquisável para times de IA

Ver no GitHub
BomPluginScore · de 0 a 100

Infraestrutura de memória aberta para equipes que usam IA. Centraliza notas, links, memórias e relacionamentos em um repositório único, tornando-os acessíveis via MCP para Claude, Cursor e outras ferramentas compatíveis. Elimina reset loops ao manter contexto persistente e reutilizável entre sessões.

Instalação

$npx convez-ai kumbukum/kumbukum

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
48
Forks
0
Acompanhando
1
Saúde
50%
Último commit
há 3 dias
Versão
202609201
Issues abertas
1
Contribuidores
2

Casos de uso

  • Manter contexto persistente entre diferentes ferramentas de IA (Claude, Cursor, ChatGPT)
  • Centralizar documentação, decisões e preferências de equipe para acesso por assistentes
  • Conectar conhecimento disperso através de grafo semântico e tags

Veredito

Kumbukum resolve um problema real: equipes de IA precisam de memória compartilhada e inspeção sobre o contexto passado aos assistentes. A implementação como MCP server é competente e integra bem com o ecossistema Claude, oferecendo tanto opção cloud quanto self-hosted. Recomendado para times que querem controlar e reutilizar conhecimento institucional.

README

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

Kumbukum

Open source memory infrastructure for teams.
Turn scattered company knowledge into AI-ready context.

Intro video

Watch the video

Stop reset loops. Give every AI tool trusted memory your team can inspect, control, and reuse.

Kumbukum is an open-source memory layer for AI-native teams and MCP-compatible tools. Store notes, memories, URLs, and relationships in one place, then let assistants retrieve the right context across sessions, projects, and clients.

Website · Cloud · Docs · Self-Hosted Guide · MCP Docs · API Reference

Kumbukum homepage preview

Why Kumbukum?

AI tools are great at reasoning and terrible at remembering.

Kumbukum gives you a shared, searchable, editable memory layer that sits between your team and your AI tools:

  • Persistent context — keep preferences, decisions, docs, and bookmarks across sessions
  • Works across tools — use the same memory store with Claude Desktop, Cursor, and other MCP clients
  • Open and inspectable — your memory is not trapped in a black box
  • Built for teams — organize knowledge by project, connect related items, and make it reusable

What’s included

  • Notes — rich-text documents with full-text and semantic search
  • Memories — short-form context for preferences, decisions, and learnings
  • URLs — saved pages with extracted content and optional full-site crawling
  • Knowledge Graph — manual, tag-based, and semantic links between items
  • AI Chat — search and manage knowledge with natural language
  • MCP Server — 44 tools for notes, memories, URLs, projects, graph, and search
  • Browser Extension — save notes and URLs from anywhere
  • Bidirectional Git Sync — sync project knowledge to and from Git repositories

How it works

  1. Capture knowledge — add notes, store memories, save URLs, or import documents
  2. Connect your tools — plug Kumbukum into Claude Desktop, Cursor, or another MCP client
  3. Retrieve the right context — your assistant searches and reuses what matters, instead of starting from zero every time

Cloud or self-hosted

Kumbukum Cloud and the self-hosted edition share the same product features. The difference is who runs the infrastructure.

CloudSelf-Hosted
FeaturesAllAll
HostingManaged by usYou manage
UpdatesAutomaticManual
BackupsIncludedYou configure
PricingSubscriptionFree (open source)

Quick start

Kumbukum Cloud

  1. Sign up at app.kumbukum.com
  2. Create your first project
  3. Generate a personal access token in Settings → Tokens
  4. Add Kumbukum to your AI tool

Example Claude Desktop config on macOS:

{
    "mcpServers": {
        "kumbukum": {
            "command": "npx",
            "args": ["-y", "mcp-remote", "https://mcp.kumbukum.com/mcp"],
            "env": {
                "ACCESS-TOKEN": "your-access-token"
            }
        }
    }
}

Self-hosted

Production with Docker Compose

Requirements:

  • Node.js >= 24
  • Docker & Docker Compose
  • pnpm 10+ (for local development)
  • MongoDB 6+
  • Redis 6+
  • Typesense 31+

Grab the production Compose file and pass configuration as shell environment variables:

curl -O https://raw.githubusercontent.com/kumbukum/kumbukum/main/compose.prod.yml

APP_URL=https://your-instance.com \
SESSION_SECRET=your-session-secret \
JWT_SECRET=your-jwt-secret \
TYPESENSE_API_KEY=your-typesense-key \
SMTP_HOST=smtp.example.com \
SMTP_USER=you@example.com \
SMTP_PASS=your-smtp-password \
SMTP_FROM=noreply@example.com \
GOOGLE_API_KEY=your-google-api-key \
docker compose -f compose.prod.yml up -d

For multiple outbound SMTP servers, set SMTP_SERVERS to a JSON array of server objects. Kumbukum sends through them round-robin.

Example:

SMTP_SERVERS='[
	{
		"name": "smtp-1",
		"host": "smtp1.example.com",
		"port": 587,
		"secure": false,
		"user": "smtp-user-1",
		"pass": "smtp-pass-1",
		"from": "server@kumbukum.com"
	},
	{
		"name": "smtp-2",
		"host": "smtp2.example.com",
		"port": 465,
		"secure": true,
		"user": "smtp-user-2",
		"pass": "smtp-pass-2",
		"from": "server@kumbukum.com"
	}
]'

In Docker Compose YAML, quote it as a single JSON string:

environment:
    SMTP_FROM: server@kumbukum.com
    SMTP_SERVERS: >-
        [{"name":"smtp-1","host":"smtp1.example.com","port":587,"secure":false,"user":"smtp-user-1","pass":"smtp-pass-1","from":"server@kumbukum.com"},{"name":"smtp-2","host":"smtp2.example.com","port":465,"secure":true,"user":"smtp-user-2","pass":"smtp-pass-2","from":"server@kumbukum.com"}]

Only host is required per server. port defaults to 587, secure defaults to true only for port 465, and from falls back to SMTP_FROM.

No .env file is required for this setup.

Default service ports:

ServicePort
App3000
WebSocket3001
MCP Server3002
MongoDB27017
Redis6379
Typesense8108

Local development

For repository development, make sure MongoDB, Redis, and Typesense are available, then run:

git clone https://github.com/kumbukum/kumbukum.git
cd kumbukum
pnpm install
pnpm dev

Useful scripts:

  • pnpm dev — run the app with auto-reload
  • pnpm build — build frontend assets
  • pnpm test — run the test suite
  • pnpm test:mcp — run MCP-specific tests
  • pnpm docs:dev — run the docs site locally

MCP server

Kumbukum includes a built-in Model Context Protocol server with:

  • 44 tools across notes, memories, URLs, projects, graph, search, and AI chat
  • Shared memory across tools so Claude Desktop, Cursor, and others pull from the same context
  • Three transports — stdio, SSE, and Streamable HTTP
  • Token-based auth via Authorization: Bearer, access-token, or stdio env vars
  • Automatic default project selection so tools work without extra setup

Self-hosted example:

env 'ACCESS-TOKEN'=your-access-token API_BASE_URL=https://your-instance.com node apps/mcp/server.js

Architecture at a glance

  • Backend — Node.js + Express
  • Database — MongoDB
  • Search — Typesense for full-text and semantic retrieval
  • Cache / real-time — Redis + Socket.IO
  • Frontend — Pug templates + vanilla JavaScript
  • MCP app — apps/mcp/ for agent integrations

Documentation

License

Licensed under AGPL-3.0.

Ver o original no GitHub

Bom para

Relacionados