MCP-Connected Development Workflow for Data-Informed Prototyping

Context

My first AI coding loop relied on describing a problem in chat, copying code into the editor, and manually adding whatever context the model was missing. That worked for isolated tasks but became fragile when a prototype depended on research, product strategy, analytics, and an existing design system.

This prototype explored the next step: connecting the model to controlled project context through Model Context Protocol rather than repeatedly reconstructing that context in prompts.

My role

I designed the workflow, selected and structured the source material, configured the MCP connections, and used the system to develop functional React prototypes inside the normal development environment.

Connected context

The workflow made several kinds of evidence available:

• Databricks, Google Analytics, and Microsoft Clarity signals.
• Survey findings and notes from customer visits.
• Strategy documents and product roadmaps.
• Figma and Storybook design references.
• External benchmarks and market research.

The sources did not have equal authority. Direct user evidence and current strategy took priority over generic external material, while the design system constrained how ideas became interfaces.

Development loop

Analytics + research + strategy + design
                    ↓
             Scoped MCP access
                    ↓
          Claude in the code editor
                    ↓
             React prototype
                    ↓
           Review and iteration

The model could inspect relevant context, propose an implementation, modify the prototype, and respond to feedback without every iteration starting from an empty prompt.

What the prototype demonstrated

• Rich context can shorten the path from product question to testable interface.
• Source priority matters as much as source access.
• Development remains a review loop; MCP improves grounding but does not remove judgment.
• The same pattern can connect agents to data and tools beyond software development.

This work marked the transition from using AI as a coding assistant to designing systems in which models could operate against live, scoped context. That principle now sits inside Company OS.

Stack

• Claude
• Model Context Protocol
• React
• Databricks and web analytics
• Azure
• VS Code

Status: Prototype and development workflow. It demonstrated the integration pattern rather than a standalone commercial product.