convez.ai
MCP ServerDados & AnalyticsAvançado
yogsoth-ai

Wiki-Vault MCP Server

yogsoth-ai/knowledge-structuring

Grafo de conhecimento persistente com busca e validação para pesquisa autônoma

Ver no GitHub
InicialPluginScore · de 0 a 100

Servidor MCP para construir e gerenciar um substrato de conhecimento estruturado como grafo tipado. Fornece busca BM25, traversal de grafos, validação em lote e indexação incremental — sem dependências de LLM. Integra-se nativamente com Claude para compilar achados de pesquisa em ontologias, modelos causais e grafos de argumentação.

Instalação

$npx convez-ai yogsoth-ai/knowledge-structuring

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
2
Forks
0
Acompanhando
0
Saúde
50%
Último commit
há 5 meses
Versão
—
Issues abertas
0
Contribuidores
1

Casos de uso

  • Construir bases de conhecimento estruturado para sistemas de pesquisa autônoma
  • Manter ontologias e grafos causais como memória persistente de agentes IA
  • Validar e auditar integridade de grafos de conhecimento em larga escala
  • Compilar achados de pesquisa em conexões tipadas e recuperáveis

Veredito

Ferramenta sólida e bem arquitetada para quem precisa de memória estruturada persistente em sistemas de IA autônomos. Exige compreensão de grafos e conhecimento de teoria, mas oferece abstração clara (9 entidades, 10 tipos de aresta, 8 ferramentas focadas). Não é para usuários casuais — é infraestrutura de pesquisa e desenvolvimento.

README

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

"The only true wisdom is in knowing you know nothing." — Socrates

📚 wiki-vault

Lightweight knowledge vault MCP server for structured wiki compilation.

🧭 Part of the De-Anthropocentric Research Engine. This repository is the knowledge-structuring component of DARE — the full autonomous research-orchestration system — providing the wiki-vault MCP server plus the skills that compile research findings into structured knowledge (ontologies, causal graphs, design-space maps, argument graphs). DARE bundles it together with eight other research packages into a single self-contained clone. For the complete engine and the recommended way to use it, see the main repository.

wiki-vault is not a note-taking app. It is a structured knowledge substrate — a typed graph of interconnected research artifacts that grows autonomously as AI agents compile findings. BM25 search, typed edges, batch validation, zero LLM dependencies. The vault is the persistent memory layer where research knowledge crystallizes into reusable structure.


⚡ What It Does

  • 🔍 BM25 full-text search — ranked retrieval across all vault pages with type/tag filters and contextual snippets
  • 🕸️ Typed knowledge graph — 10 edge types connecting 9 entity types, with BFS traversal and orphan detection
  • 🩺 Batch validation — 5 check types (broken links, orphans, missing frontmatter, duplicate edges, stale index) with auto-fix
  • 📊 Graph statistics — global metrics (density, orphans, type distribution) and per-node connectivity analysis
  • 🔄 Incremental indexing — only re-indexes changed files, full rebuild available on demand
  • 📖 Obsidian-compatible — standard markdown + YAML frontmatter + [[wikilinks]], browsable without modification

🎯 Design Philosophy

  • Single unified vault. All knowledge in one place regardless of research domain. Cross-domain connections emerge naturally from the graph.
  • CC handles CRUD directly. File creation, editing, deletion are CC's native operations. The MCP server provides only what CC cannot do efficiently: ranked search, graph traversal, batch validation.
  • Sources are immutable. Raw material is preserved verbatim. Synthesis and evolution happen in wiki pages.
  • Graph is first-class. Every page connects via typed edges. Orphans are failures, not features.
  • 兵法书, not pipeline. Skills teach principles and provide SOPs — CC decides strategy and execution autonomously.

🏗️ Architecture

┌─────────────────────────────────────────────────────────┐
│  MCP Server (8 tools)                                   │
│  vault_search · vault_add_edge · vault_query_graph      │
│  vault_graph_stats · vault_lint · vault_index           │
│  vault_info · vault_edge_audit                          │
├─────────────────────────────────────────────────────────┤
│  Core Modules                                           │
│  index.ts (BM25) · graph.ts (edges+traversal) · lint.ts│
├─────────────────────────────────────────────────────────┤
│  Vault (filesystem)                                     │
│  9 entity dirs · _edges.jsonl · _index.json · _schema.md│
└─────────────────────────────────────────────────────────┘

Tools

ToolDescription
vault_searchBM25 full-text search with type/tag filters and snippets
vault_add_edgeCreate typed edge between pages (validates, deduplicates)
vault_query_graphBFS traversal from node with direction/depth/type filters
vault_graph_statsGlobal or per-node graph statistics and orphan detection
vault_lintBatch validation (5 check types) with optional auto-fix
vault_indexRebuild search index (incremental or full)
vault_infoReturns vault metadata (root, directories, types, stats) for session init
vault_edge_auditScan edges and report missing [[wikilinks]] in source page bodies

Skills (9)

LevelSkillPurpose
Tacticknowledge-compilationCompile research findings into vault pages
Tacticvault-maintenanceVault health upkeep and cleanup
SOPwiki-searchSearch before creating (deduplication)
SOPwiki-graph-queryExplore node neighborhood
SOPwiki-add-edgeCreate typed relationships
SOPwiki-ingest-sourceWrite immutable source pages
SOPwiki-compile-pageCreate/update synthesized pages
SOPwiki-lint-fixRun lint and optionally fix
SOPwiki-edge-auditAudit and fix missing wikilinks for edges

Entity Types

TypeDirectoryPurpose
sourcesources/Immutable raw material
conceptconcepts/Synthesized understanding
entityentities/Named entities (people, orgs, tools)
claimclaims/Testable assertions with confidence
relationrelations/Named relationships
questionquestions/Open research questions
evidenceevidence/Supporting/refuting evidence
failurefailures/Documented failed approaches
topictopics/High-level topic containers

Edge Types

Edge TypeSemantics
component_ofA is part of B
instance_ofA is an example of B
supported_byA has evidence from B
contradictsA conflicts with B
supersedesA replaces B
derived_fromA was built from B
addressesA answers/solves B
raisesA creates/implies B
failed_forA didn't work for B
related_toWeak association

🚀 Quick Start

# Install
npm install

# Run tests (83 tests)
npm test

# Start MCP server
VAULT_ROOT=/path/to/vault npx @yogsoth-ai/wiki-vault

MCP Configuration

Add to your .mcp.json:

{
  "wiki-vault": {
    "type": "stdio",
    "command": "npx",
    "args": ["@yogsoth-ai/wiki-vault"],
    "env": { "VAULT_ROOT": "/path/to/your/vault" }
  }
}

📄 License

Apache-2.0


A component of the De-Anthropocentric Research Engine, part of the Yogsoth AI ecosystem. Built by Pthahnix.

Ver o original no GitHub

Bom para

Relacionados