Multi-Agent Workflow: Startup MVP with Persistent Memory

The same startup MVP workflow from workflow-startup-mvp.md, but with an MCP memory server handling state between agents. No more copy-paste handoffs.

The Problem with Manual Handoffs

In the standard workflow, every agent-to-agent transition looks like this:

Activate Backend Architect.

Here's our sprint plan: [paste Sprint Prioritizer output]
Here's our research brief: [paste UX Researcher output]

Design the API and database schema for RetroBoard.
...

You are the glue. You copy-paste outputs between agents, keep track of what's been done, and hope you don't lose context along the way. It works for small projects, but it falls apart when:

• Sessions time out and you lose the output
• Multiple agents need the same context
• QA fails and you need to rewind to a previous state
• The project spans days or weeks across many sessions