Mark Korotin
AI-native Framework

DjangoCFG

The framework that disappears so your product can appear.

djangocfg.com
DjangoCFG — The framework that disappears so your product can appear.
Problem

Standing up a real SaaS backend — config, admin, payments, realtime, docs — takes weeks of boilerplate, and the result is opaque to the AI coding agents teams now build with.

Insight

The framework should disappear, and it should be legible to machines. Wire the proven pieces together once behind a type-safe configuration layer, and expose the whole capability graph through MCP so coding agents query what it can do instead of scraping prose docs.

System

DjangoCFG is an opinionated, AI-native framework that replaces Django's settings model with a type-safe Pydantic v2 system and unifies Django, a Next.js admin, payments, and realtime (Centrifugo) into one coherent architecture. Capabilities are exposed through an MCP server, so agents read schemas directly.

Architecture

A single configuration object drives the stack: models validated through Pydantic v2, MCP endpoints generated from type definitions, a Next.js frontend scaffolded to match, and a realtime layer wired into the Django event bus. Convention over configuration — but the conventions are right.

Framework decisions are architectural decisions made once for every team that uses them. DjangoCFG is opinionated because indecision at the infrastructure layer creates permanent technical debt for every product built on top of it.

The Hard Part

Making MCP a first-class citizen of a Django application required rethinking how Django exposes its internal model graph. The result is a type-safe bridge between Django ORM definitions and MCP tool specifications — generated, not handwritten.

Impact
First-deploy setup
Weeks→30s
First-deploy setup
Native AI interface
MCP
Native AI interface
Open source
OSS
Open source
Django 5.2+Pydantic v2MCPNext.jsCentrifugogRPCRedisPostgreSQL
What's Next

Native support for agent-facing APIs, streaming responses, and a plugin model that lets teams extend the configuration layer without forking the framework.