Kratos MCP

Ultra-Lean Memory System for AI Coding Tools

Never explain your codebase again. Let AI remember everything.

Kratos has evolved. This repo is now legacy.

github.com/ceorkm/kratos-cli

Kratos is now a standalone CLI — faster, lighter, works with ANY AI agent.

No MCP overhead. No vendor lock-in. Just memory that works.

npx kratos-memory

Claude Code, Codex, Cursor, Cline — anything that runs Bash.

Your existing memories work instantly. Same ~/.kratos/ storage, same SQLite format.

Go to Kratos CLI >

Why Kratos?

After building 30+ production apps with AI, we discovered a critical problem: AI tools forget everything between sessions. You explain your architecture, your patterns, your decisions—and tomorrow, you explain it all again.

Kratos MCP solves this with an ultra-lean memory system that gives AI perfect recall of your project—with minimal context overhead.

Features

100% Project Isolation

Each project gets its own SQLite database. No cross-contamination. Ever.

Zero Configuration

Auto-detects projects via git, package.json, or directory structure. Just install and code.

Ultra-Lean Architecture

Just 12 essential tools. 64% smaller context footprint than competitors.

Universal Protocol

Works with Claude, Cursor, Windsurf, Continue—any MCP-compatible tool.

Installation

# Install globally
npm install -g kratos-mcp

# Or run directly with npx (no installation required)
npx kratos-mcp

# Or install as a dependency
npm install kratos-mcp

Quick Start

1. Configure Your AI Tool

Add to your Claude Desktop config file:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json
• Linux: ~/.config/claude/claude_desktop_config.json

{
  "mcpServers": {
    "kratos": {
      "command": "npx",
      "args": ["--yes", "kratos-mcp@latest"]
    }
  }
}

Or if you have it installed globally:

{
  "mcpServers": {
    "kratos": {
      "command": "kratos-mcp",
      "args": []
    }
  }
}

Run this command in your terminal:

claude mcp add kratos -- npx --yes kratos-mcp@latest

Or for global installation:

# First install globally
npm install -g kratos-mcp@latest

# Then add to Claude Code
claude mcp add kratos -- kratos-mcp

Add to .cursor/mcp_config.json in your project root:

{
  "mcpServers": {
    "kratos": {
      "command": "npx",
      "args": ["--yes", "kratos-mcp@latest"]
    }
  }
}

Kratos works with any tool supporting the Model Context Protocol. The general format is:

{
  "command": "npx",
  "args": ["kratos-mcp"]
}

Compatible with: Windsurf, Cline, BoltAI, Augment Code, Roo Code, Zencoder, Amazon Q, Qodo Gen, JetBrains AI, Warp, Opencode, Continue.dev, Zed, and more!

Check your tool's documentation for specific MCP server configuration location.

2. Start Using Kratos

// Your AI now remembers:
// - Your authentication patterns
// - Your API structure
// - Your component architecture
// - Your coding standards
// - Every decision you've made

Available Tools

Kratos provides 12 ultra-lean tools optimized for minimal context consumption:

Memory Management (7 tools)

| Tool | Description | |------|-------------| | memory_save | Store important project knowledge with tags, paths, and importance levels | | memory_search | Smart semantic search with debug mode and path matching | | memory_ask | Natural language queries about your memories | | memory_get_recent | Get recently created memories with filtering | | memory_get | Retrieve a specific memory by ID | | memory_get_multiple | Bulk retrieve multiple memories | | memory_forget | Delete a memory by ID |

Security (1 tool)

| Tool | Description | |------|-------------| | security_scan | Scan text for PII and secrets before saving |

Project Management (3 tools)

| Tool | Description | |------|-------------| | project_switch | Switch between different projects | | project_current | Get current active project info | | change_storage_path | Dynamically change storage location with automatic data migration |

System (1 tool)

| Tool | Description | |------|-------------| | system_status | Get system status and memory statistics |

How It Works

AI Tool ──► Kratos MCP ──► Project Detection ──► SQLite + FTS5 ──► Perfect Context

• SQLite + FTS5: Lightning-fast full-text search
• Smart Scoring: Path matching + recency + importance
• Auto-detection: Git, package.json, or directory-based
• Secure: All data stays local, no external calls
• Lean: Only 4 core components, minimal memory footprint

Performance

| Metric | Value | |--------|-------| | Context Overhead | 64% smaller than v3 | | Memory Retrieval | < 10ms | | Project Switch | < 100ms | | Storage Overhead | ~2MB per project |

Data Storage

~/.kratos/
├── projects/
│   ├── project-id-1/
│   │   └── memories.db          # SQLite database with FTS5
│   └── project-id-2/
│       └── memories.db
└── global/
    └── global.db                # Shared knowledge (optional)

Use change_storage_path to move data to custom locations like /opt/kratos or .kratos for per-project storage.

Example Usage

// Save a memory
await memory_save({
  summary: "JWT auth implementation",
  text: "We use httpOnly cookies with refresh tokens...",
  tags: ["auth", "security"],
  paths: ["src/middleware/auth.ts"],
  importance: 5
});

// Search memories
await memory_search({
  q: "authentication",
  k: 5,
  debug: true
});

// Ask natural language questions
await memory_ask({
  question: "How does our auth system work?",
  limit: 10
});

Contributing

git clone https://github.com/ceorkm/kratos-mcp.git
npm install
npm run build
npm run dev

License

MIT

Built for developers who value their time.

Report Bug · Request Feature