Everyone says, "Talk to users" and "Test your hypotheses," but when you're deep in the code, the urge to just build the thing is hard to resist.
So I did something unusual. I turned an entire product discovery framework into a Model Context Protocol (MCP). Not a chatbot. Not a library of templates. A full protocol that my AI helper uses to guide me through discovery, even when I'm reluctant.
What Makes It Different
What makes it different is how comprehensive it is. The Discovery Flywheel MCP includes nine tools that walk through a complete discovery loop: outcomes, personas, problems, hypotheses, solutions, experiments, prioritization, and risk.
When I start a new project, I run discovery_scaffold. That spins up a workspace with preloaded research tools, ready-to-fill templates, and a README that marks each stage of the process.
Then the questions start.
- "If this works, what user behavior will change?"
- "What's your riskiest assumption?"
- "How can you be sure this hypothesis is wrong?"
It's like pair programming with a product coach who won't let me get away with hand-waving.
Even When I Just Want to Build
Even when I just want to build, the system pushes me to pause, think, and validate.
And it's working. I spend less time building things no one asked for, and more time understanding the people I'm actually building for.
At Curebase, we've been running this approach for the last four months, and the difference in our trial orchestration product is night and day:
