Drift

Stop your AI from writing code that doesn't match your codebase.

Drift learns the patterns in your code and gives AI agents the context they need to write consistent code. No more "follow the patterns in this file" prompts that get ignored. No more 10k token steering documents that go stale.

The Problem

You've got patterns in your codebase. Error handling, API responses, component structure, auth checks - stuff that took months to establish. Then you ask an AI to add a feature and it writes something completely different. Not wrong, just... not yours.

You can paste examples. You can write massive prompt documents. But the AI doesn't really know your patterns - it's just guessing based on whatever context fits in the window.

What Drift Does

1. Scans your codebase with 101 detectors across 15 categories
2. Learns your patterns - not just syntax, but semantic meaning (how you handle errors, structure APIs, etc.)
3. Exposes patterns to AI via MCP so agents can query what they need
4. Detects FE/BE contract mismatches before they become runtime bugs
5. Flags violations when code drifts from established patterns

Quick Start

# Install the CLI
npm install -g driftdetect

# Initialize in your project
cd your-project
drift init

# Scan for patterns
drift scan

# See what it found
drift status

That's it. You now have a .drift/ folder with all your learned patterns.

For AI Agents (MCP)

This is where it gets good. Install the MCP server:

npm install -g driftdetect-mcp

Add to your MCP config (Kiro, Claude Desktop, Cursor, etc.):

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

Now your AI can query patterns instead of guessing:

Tool	What it does
drift_status	Pattern health overview
drift_patterns	Query by category with confidence scores
drift_files	Patterns in specific files
drift_where	Find where patterns are used
drift_contracts	FE/BE API mismatches
drift_examples	Real code examples from your codebase
drift_pack	Pre-built pattern bundles for common tasks
drift_export	AI-optimized pattern export
drift_feedback	Rate examples to improve suggestions

Pattern Categories

Drift detects patterns across 15 categories:

Category	What it learns
structural	File naming, imports, module boundaries, barrel exports
components	Props patterns, hooks usage, composition, render patterns
styling	CSS methodology, spacing scales, color tokens, responsive patterns
api	Route structure, response envelopes, HTTP methods, versioning
auth	Token handling, RBAC patterns, session management
errors	Exception hierarchy, error codes, try/catch patterns
data-access	Repository patterns, DTOs, query builders
testing	Test structure, mocking patterns, fixtures
logging	Log levels, structured logging, correlation IDs
security	Input validation, sanitization, secrets handling
config	Environment variables, feature flags
types	Interface vs type usage, generics, utility types
performance	Memoization, lazy loading, debounce/throttle
accessibility	ARIA patterns, keyboard navigation, focus management
documentation	JSDoc style, README patterns, code comments

Contract Detection

Drift finds mismatches between your frontend API calls and backend endpoints:

GET /api/v1/billing/subscription
  Backend returns: { plan_name: string | null, status: string | null }
  Frontend expects: { plan_name: string, status: string }
  
  ⚠️ Nullability mismatch - frontend doesn't handle null

No more "works on my machine" because the backend added an optional field.

CLI Reference

See docs/cli-reference.md for full command documentation.

Links

• npm: driftdetect - CLI
• npm: driftdetect-mcp - MCP Server
• npm: driftdetect-core - Core engine
• npm: driftdetect-detectors - Pattern detectors

License

MIT