Flow of launching MVP with Kiro and Claude for product managers

Launching an MVP Without a Team: A Product Manager's Flow with Claude, Kiro, and MCP

A product manager without deep technical expertise can build and deploy an MVP in a week. Testing has shown this works for a personal consulting website and an agile tool as a Jira alternative for startups. The AI-powered stack minimizes manual coding, focusing on discovery and automation.

The key is a predictable process: from systematic prompting to deployment on GitHub Pages. Saves up to 40 hours on analysis and design.

Choosing a Stack for Vibe-Coding

The stack is designed for seamless stage transitions without losing context:

| Tool | Role |

|------------|------|

| Claude Sonnet 4.5 | Discovery, systematic prompts, analytics |

| Kiro (Amazon) | Architecture assembly, implementation, deployment |

| Figma Make | Frontend generation from prompts |

| Supabase | Backend with database |

| GitHub Pages | Static site hosting |

| Context7 MCP | Session context management |

| Playwright MCP | End-to-end testing |

This combination replaces the roles of analyst, designer, and developer for pilot projects.

Flow: A Detailed Breakdown of Stages

Discovery with Claude Sonnet 4.5

Create a project assistant with a system prompt: specify audience, pain points, Jobs-to-be-Done, business value, constraints. Output includes:

• Ready-made analytics and content plan.
• Design prompt.
• Architectural prompt (optional for review).

For an MVP, architecture isn't critical—focus on the hypothesis.

Frontend Generation with Figma Make

Pass artifacts from Claude to Figma Make. Download ready-made files without modifications. Stable results for pilots: interfaces cover core functionality.

Build and Architecture with Kiro

Upload frontend + analytics. Kiro acts as architect + developer:

• Documents decisions in text.
• Implements code.
• Requires justification for each step.

The process is controlled: the product manager sees the logic without reading source code.

Deployment with MCP Servers

Install MCP via Kiro:

• Context7 MCP: Saves context between sessions, prevents loss of project knowledge.
• Supabase MCP: Auto-generates tables, schemas, migrations.
• Playwright MCP: Browser automation, basic end-to-end testing.

Deployment steps:

• Set up GitHub/Supabase accounts.
• Kiro creates the database and deploys to Pages.
• Generates instructions for custom domains.

The MVP is ready for validation.

Comparing Tool Combinations

Optimal: Kiro + Claude Sonnet 4.5. Minimal context switching, full artifact connectivity.

Alternative: GPT + Figma Make + Kiro. Figma Make is stable on the frontend regardless of upstream.

| Combination | Stability | Connectivity |

|--------|--------------|-----------|

| Kiro + Claude | High | Maximum |

| GPT + Figma + Kiro | Medium | Good |

Practical Limitations

This flow is for MVPs, not production:

• Requires technical architecture review for load handling.
• Context7 doesn't prevent context window overflow in large projects.
• Playwright MCP covers happy paths; edge cases require manual handling.
• Kiro may choose suboptimal architecture without review.

Key Takeaways

• Economics: Launch without hiring, validate hypotheses in a week.
• Control: Tools document decisions in writing.
• Scale: MVP before building a team, not a replacement for developers.
• Stability: Kiro+Claude minimizes iterations.
• Context: MCP servers solve state loss issues.

This approach lowers the entry barrier: requires a theoretical overview of dev processes and prompting skills.

— Editorial Team