WeVibe MCP — Gateway Local de Memória e Identidade
WeVibe-Network/wevibe-mcp
Gateway local de recall, segurança e aprovação humana para memória distribuída
Ver no GitHubServidor MCP TypeScript que implementa camada local de segurança, recall de memória e aprovação humana para agentes de IA. Gerencia recuperação criptografada de memória, aplicação de políticas locais, sanitização de artefatos e controle de identidade sob ~/.wevibe/.
Instalação
npx convez-ai WeVibe-Network/wevibe-mcp1 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
- Integrar agentes de IA com memória persistente e criptografada na WeVibe Network
- Aplicar políticas de segurança e controle humano antes de injetar memória em sessões ativas
- Recuperar e descriptografar localmente candidatos de memória com validação de guard
Veredito

Componente maduro (alpha) do ecossistema WeVibe, bem-arquitetado para segurança local e recall inteligente. Exige familiaridade com MCP, criptografia e infraestrutura de agentes, adequado para integradores e pesquisadores de memória distribuída.
README
A tradução para o português deste README ainda está na fila.
Local safety, identity, and approval gateway for WeVibe memory in coding agents.
Overview
wevibe-mcp is a TypeScript/Node (ESM) service that runs on the developer's machine and handles local memory retrieval and policy enforcement.
Core responsibilities in the current implementation:
- Runs an MCP server and a loopback HTTP API (
127.0.0.1:4450, Bearer-token gated). - Builds recall queries from local session signals and computes query embeddings locally.
- Fetches encrypted memory candidates from the hub, then decrypts them locally through the Umbral sidecar flow.
- Applies local safety controls (
wevibe-guardscan, artifact policy checks, OCR sanitization, blacklist handling) before candidate presentation. - Powers the human approval gate used before memory injection.
- Maintains local identity and keystore data under
~/.wevibe/. - Ships CLI entrypoints:
wevibe-mcp,wevibe-admin, andwevibe-retrieve. - Bundles OpenCode plugin source for local onboarding automation.
Note: this repository still contains some legacy Python-era artifacts, but the shipped runtime path is TypeScript.
Role in the WeVibe Network
wevibe-mcp is the local enforcement layer between an agent client and network memory.
It is responsible for:
- Local recall query shaping and embedding.
- Secure retrieval + local decrypt path.
- Local guard/policy enforcement.
- Mandatory human review before memory reaches the active coding session.
In practice, this is the path that governs memory injection into the agent runtime.
Getting started
Prerequisites
- Node.js
>=20 - npm
- Local WeVibe dependencies used by your setup (for example,
wevibe-guard, Umbral sidecar binary, and model provider such as Ollama)
Build
npm install
npm run build
Run
Development mode:
npm run dev
Run built artifacts:
node dist/admin.js setup-identity
node dist/server.js
Optional OpenCode onboarding helper:
node dist/admin.js install-opencode
Docker support is available via the repository Dockerfile.
Testing
npm test
npm run test:integration
Configuration
Common environment variables:
WEVIBE_HUB_URL— hub API base URL.WEVIBE_CHAIN_REST_URL— chain REST endpoint used for resolver workflows.WEVIBE_DASHBOARD_URL— dashboard base URL used by pairing/adoption flows.WEVIBE_OLLAMA_URL/OLLAMA_HOST— local model + embedding endpoints.WEVIBE_EXTRACTION_MODEL— extraction model ID.WEVIBE_EMBEDDING_MODEL— embedding model ID.WEVIBE_HTTP_HOST— bind host for local HTTP API (default loopback).WEVIBE_DASHBOARD_PORT— local dashboard server port.WEVIBE_UMBRAL_SIDECAR_BIN— path to Umbral sidecar binary.
Local files/paths to be aware of:
~/.wevibe/mcp-session-token— token consumed by local clients for HTTP auth.~/.wevibe/plugin-config.json— local risk/provider policy config.~/.wevibe/— identity, keystore, and local state.
Roadmap
See ROADMAP.md.
License
Apache-2.0.