Kernal MCP server

Kernal

Open-source knowledge graph for professionals. Auto-extracts entities and relationships from natural conversation via MCP.

Talk to Claude naturally about your meetings, calls, and interactions. Kernal stores people, organizations, topics, and relationships — building a knowledge graph you own.

What's Included (Open Source)

Everything you need to run Kernal locally on your own machine:

• 13 MCP tools — ingestion, CRUD, query, corrections (see full list below)
• SQLite database — local-first, your data never leaves your machine
• LLM-driven extraction — Claude reads your text, decides what to extract, and calls structured write tools
• Entity resolution — fuzzy matching + Levenshtein distance prevents duplicates
• CLI — init, serve, status, export
• Cloud server — Express.js with API key auth, rate limiting, CORS, session management
• Dashboard — React app with network graph, timeline, action items, overview
• 50 tests — comprehensive test suite

This is a fully functional knowledge graph you can run yourself, for free, forever.

What Andes Provides (Managed Service)

For teams and professionals who want more, Andes offers:

• Cloud hosting — access your knowledge graph from any device, no self-hosting
• Dashboard — hosted interactive visualizations powered by your data
• Multi-user — team features, shared knowledge bases, role-based access
• Onboarding & support — we set it up for you and help your team get value from day one
• Industry workflows — pre-built patterns for executive search, consulting, professional services

The open-source core is the engine. Andes wraps it with infrastructure, UX, and support.

Quick Start

npx kernal-mcp init

This creates a SQLite database at ~/.kernal/kernal.db and prints the config to add to Claude Desktop.

Add to your claudedesktopconfig.json:

{
  "mcpServers": {
    "kernal": {
      "command": "npx",
      "args": ["-y", "kernal-mcp", "serve"]
    }
  }
}

Restart Claude Desktop. Then talk naturally:

"I had lunch with Jonas Lindberg from Nordvik Energy today. He's their VP of Digital. We discussed their cloud migration — targeting Q3."

Claude extracts Jonas, Nordvik Energy, the cloud migration topic, and stores them via Kernal's write tools. Then ask:

• "What do I know about Nordvik Energy?" → Full briefing with people, interactions, topics
• "Who should I follow up with?" → Open action items with owners and due dates
• "Show me everyone at Nordvik Energy" → Contact list filtered by organization

How It Works

Kernal uses an LLM-driven extraction pattern:

1. You tell Claude about a meeting, call, or interaction
2. Claude calls kernal_remember with the raw text
3. Kernal stores the text as a note and returns extraction instructions + existing entities (for dedup)
4. Claude reads the text intelligently and calls structured write tools (kernaladdperson, kernaladdorg, kernaladdactivity, etc.)
5. Each write goes through entity resolution to prevent duplicates
6. The LLM makes all extraction decisions — no regex guessing

The MCP server is a clean data store. The LLM is the brain.

MCP Tools

Ingestion (write)

Query (read)

Shortened. The full README is on GitHub.