wiki-nexus

Zero-server knowledge graph engine for markdown wikis. Indexes pages, wikilinks, tags, topic clusters, and knowledge flows into SQLite, then exposes them through MCP tools and a REST API.

The core principle: precomputed relational intelligence. All clustering, relationship mapping, staleness detection, and impact analysis is computed at index time — not at query time via expensive LLM multi-hop. Even smaller models get complete structural context in a single tool call.

Install

npm install -g wiki-nexus

Or use without installing:

npx wiki-nexus analyze

Quick start

# Index your wiki
cd ~/my-knowledge-base
wiki-nexus analyze

# See what was built
wiki-nexus status
# → 52 pages, 187 edges, 6 clusters, 0 broken links

# Search
wiki-nexus query "compression methods"

# Hook up to Claude Code
wiki-nexus setup

# Start HTTP API
wiki-nexus serve

CLI commands

| Command | Description | |---------|-------------| | wiki-nexus analyze [path] | Index a wiki directory | | wiki-nexus status | Show index status | | wiki-nexus query <query> | Search the wiki | | wiki-nexus list | List all indexed wikis | | wiki-nexus clean | Delete index | | wiki-nexus mcp | Start MCP server (stdio) | | wiki-nexus serve | Start HTTP API server | | wiki-nexus setup | Configure MCP for your editor |

Flags

wiki-nexus analyze --force             # Full re-index
wiki-nexus analyze --embeddings        # Enable semantic search
wiki-nexus analyze --index-dir ./data  # Custom index location
wiki-nexus serve --port 8080           # Custom port
wiki-nexus query "topic" --limit 5     # Limit results
wiki-nexus clean --all --force         # Delete all indexes

MCP tools

Once indexed, the MCP server exposes 8 tools for AI agents:

| Tool | What it returns | |------|----------------| | wiki_query | Hybrid search results (BM25 + semantic + RRF fusion) | | wiki_context | 360-degree page view — all relationships, clusters, flows, staleness | | wiki_impact | Blast radius — what pages are affected if this page changes | | wiki_clusters | Topic clusters with members and cohesion scores | | wiki_lint | Health check — broken links, orphans, stale pages | | wiki_flows | Knowledge flow tracing from sources through derived pages | | wiki_stats | Wiki-level statistics | | list_wikis | All indexed wikis |

Editor setup

wiki-nexus setup  # Auto-detects Claude Code, Cursor, Codex, Claw Code

Or manually:

Claude Code:

claude mcp add wiki-nexus -- npx wiki-nexus mcp

Cursor (~/.cursor/mcp.json):

{
  "mcpServers": {
    "wiki-nexus": {
      "command": "npx",
      "args": ["wiki-nexus", "mcp"]
    }
  }
}

HTTP API

wiki-nexus serve

| Endpoint | Maps to | |----------|---------| | GET /api/wikis | list_wikis | | GET /api/wikis/:name/stats | wiki_stats | | GET /api/wikis/:name/query?q=... | wiki_query | | GET /api/wikis/:name/pages/:slug/context | wiki_context | | GET /api/wikis/:name/pages/:slug/impact | wiki_impact | | GET /api/wikis/:name/clusters | wiki_clusters | | GET /api/wikis/:name/lint | wiki_lint | | GET /api/wikis/:name/flows | wiki_flows |

Wiki format

WikiNexus works with markdown wikis that use:

• YAML frontmatter with title, tags, sources, updated
• [[slug]] wikilinks where slug = filename without .md
• Flat structure in wiki/pages/
• SCHEMA.md at wiki root (optional, helps detection)

How it works

1. Parse — extracts frontmatter, wikilinks, headings from every .md file
2. Build graph — pages become nodes, wikilinks become edges, stored in SQLite
3. Cluster — Louvain community detection groups pages into topic clusters
4. Trace flows — follows how knowledge flows from source documents through derived pages
5. Analyze — detects broken links, orphan pages, stale content, missing backlinks
6. Search — BM25 keyword search + optional semantic embeddings with RRF fusion

All computed at index time. MCP tools just read precomputed results.

Packages

| Package | Purpose | |---------|---------| | wiki-nexus | CLI binary + HTTP API | | @wiki-nexus/core | Core indexing engine | | @wiki-nexus/mcp-server | MCP stdio server |

License

MIT