How we solved the AI pricing paradox by building a Dual BYOK + MCP Subscription Architecture

Hey Indie Hackers!

When monetizing AI features in an open-core SaaS product, you quickly hit the "AI Pricing Paradox":

If you charge a flat subscription for AI usage, heavy power users eat up your API margins.

If you charge per-token markups, developers hate the artificial markup and refuse to adopt it.

While building NextBlock CMS, we wanted to support both casual non-technical site managers and hardcore developers who live inside VS Code or Cursor.

Our Solution: The Dual Architecture
We evolved our Cortex AI package into two distinct execution pathways:

In-Browser BYOK (Bring Your Own Key): Dashboard users paste their OpenRouter API keys into settings. Compute costs go straight to OpenRouter with zero markup from us.

MCP Gateway Subscription: For power users who want to connect Claude, ChatGPT, or Gemini directly to their live site or localhost, we host/expose an MCP (Model Context Protocol) gateway server. They pay a simple NextBlock subscription for the orchestration capabilities, while leveraging their existing AI client subscriptions for token compute!

Why This Wins
Zero Margin Risk: We don't absorb runaway token costs.

Interoperability: Users aren't locked into our UI—they can audit inventory, generate landing pages, or tweak site styling directly from their terminal or IDE.

What do you think of this monetization & architecture model for open-core AI features? Would love to hear how other bootstrappers are handling MCP integration!

⚡️ Test the Sandbox: https://nextblock.dev