Microsoft Store MCP Server with Weaver Integration

An MCP (Model Context Protocol) server for Microsoft Store submissions that includes the Weaver retrospective learning system. This server enforces step-by-step agent interactions inspired by BMAD patterns to prevent "information dumping" and ensure thoughtful dialogue.

Key Features

🎯 Core Capabilities

• Step-by-step enforcement: Agents ask ONE question at a time, wait for answers
• Weaver Review: Retrospective analysis of project patterns and lessons
• Weaver Investigation: Deep dive into failure patterns and time sinks
• Deployment Analysis: Check current deployment status and strategies
• Branch Analysis: Map branch purposes and deployment methods
• Pattern Detection: Find failure clusters and circular development

🧠 BMAD-Inspired Patterns

• Activation Sequence: Agents follow startup protocol (read instructions → load config → greet → halt)
• Mandatory Interaction: Interactive mode requires user responses between questions
• Violation Detection: Prevents agents from dumping everything at once
• Mode Selection: Choose between interactive (step-by-step) or YOLO (all at once)

Installation

# Clone the repository
git clone <repository-url>
cd microsoft-store-mcp

# Install dependencies
npm install

# Build the TypeScript
npm run build

# Run in development mode
npm run dev

Usage

Starting a Weaver Review

// Interactive mode - asks questions one at a time
{
  "tool": "weaver_review",
  "arguments": {
    "mode": "interactive"
  }
}

// YOLO mode - complete analysis without interaction
{
  "tool": "weaver_review", 
  "arguments": {
    "mode": "yolo"
  }
}

Starting an Investigation

{
  "tool": "weaver_investigate",
  "arguments": {
    "pattern": "build_failures"  // or "circular_commits", "time_sinks", "abandoned_work"
  }
}

Answering Questions

{
  "tool": "answer_question",
  "arguments": {
    "answer": "The build kept failing because we didn't understand the MSIX packaging requirements..."
  }
}

Project Structure

microsoft-store-mcp/
├── src/
│   ├── index.ts                          # Main MCP server
│   ├── weaver/
│   │   ├── orchestrator.ts               # Enforces step-by-step flow
│   │   ├── review-orchestrator.ts        # Retrospective analysis
│   │   ├── investigation-orchestrator.ts # Failure pattern investigation
│   │   └── state-manager.ts              # Tracks conversation state
│   ├── tools/
│   │   ├── deployment-analyzer.ts        # Analyzes deployments
│   │   ├── branch-checker.ts             # Maps branches
│   │   └── git-pattern-detector.ts       # Finds patterns
│   └── resources/
│       ├── weave-loader.ts               # Loads .weave files
│       └── agent-instructions.ts         # Loads agent configs
├── .weave/                                # Weaver configuration (in projects)
│   ├── config.yaml                       # Project settings
│   ├── connections.yaml                  # Human-AI collaboration log
│   └── evolution-log.md                  # Lessons learned
├── agents/                                # Agent instruction files
│   ├── weaver-review.md                  # Review agent spec
│   └── weaver-investigate.md             # Investigation agent spec
└── README.md                              # This file

How It Works

1. Interactive Review Mode

When started in interactive mode, the Weaver Review agent:

1. Deep Analysis Phase (5-10 minutes)

   • Reads all .weave/ files
   • Analyzes git history for patterns
   • Applies multiple professional perspectives
   • Synthesizes findings

2. Question Generation

   • Questions emerge from actual analysis (not scripted)
   • Prioritized by impact
   • Limited to 10-15 for reasonable dialogue

3. Step-by-Step Dialogue

   • Presents ONE question
   • Waits for answer via answer_question tool
   • Records answer
   • Presents next question
   • Continues until complete

4. Report Generation

   • Creates evolution-log.md
   • Documents patterns and lessons
   • Generates framework improvements

2. Investigation Mode

The Investigation agent focuses on undocumented problems:

1. Pattern Detection

   • Build failure clusters (5+ failures in same module)
   • Circular commits (code added, removed, re-added)
   • Time sinks (features taking 3x expected time)
   • Abandoned branches (work started but never merged)

2. Targeted Questions

   • "What really happened here?"
   • "What caused this back-and-forth?"
   • "What was the hidden complexity?"

3. Lesson Extraction

   • Documents the real constraints
   • Captures undocumented requirements
   • Identifies process improvements

Key Differences from Traditional Agents

❌ Traditional Agent Behavior

• Analyzes everything
• Asks 20 questions at once
• User overwhelmed
• Important insights lost
• No enforcement of interaction

✅ This MCP Server

• Analyzes thoughtfully
• Asks ONE question at a time
• User engaged in dialogue
• Insights captured properly
• Enforced step-by-step process

Integration with Claude Desktop

Add to your Claude Desktop configuration:

{
  "mcpServers": {
    "microsoft-store": {
      "command": "node",
      "args": ["path/to/microsoft-store-mcp/dist/index.js"],
      "env": {}
    }
  }
}

Philosophy

This MCP server embodies key lessons learned:

1. "Zealous Liar" Prevention: Agents must verify before claiming
2. 3-Failure Stop Rule: Prevent exhaustion spirals
3. One Question At A Time: Deep dialogue over information dumping
4. Retrospective Learning: Turn mistakes into system improvements
5. Investigation Mode: Uncover undocumented problems

The Weaver Philosophy

The Weaver component is not about perfection during development. It's about:

• Learning from imperfection
• Turning mistakes into system improvements
• Sharing lessons to help others
• Evolving human-AI collaboration

Example Session

User: Start a weaver review in interactive mode

MCP: ## Weaver Review Started

I've analyzed your repository and found several interesting patterns...

### Question 1 of 12:
I noticed 23 commits related to MSIX packaging over 3 days. What was the root cause that led to this pattern?

User: The documentation said one thing but the actual requirements were different. We kept trying what the docs said.

MCP: ### Captured:
Answer recorded: The documentation said one thing but...

### Question 2 of 12:
There's a circular pattern where you switched between Electron and PWA approaches 4 times. What constraint or misunderstanding caused this back-and-forth?

User: We didn't realize PWABuilder.com could handle everything. We thought we needed Electron for Store submissions.

[... continues one question at a time ...]

MCP: ## Review Complete

Evolution log generated with 12 insights captured.
Key learnings documented for future projects.

Benefits

1. Better Learning Capture: Thoughtful questions lead to meaningful answers
2. Reduced Cognitive Load: One question at a time prevents overwhelm
3. Actionable Insights: Lessons turn into concrete improvements
4. Pattern Recognition: Identifies systemic issues not just symptoms
5. Knowledge Preservation: Captures why decisions were made

Future Enhancements

• Integration with GitHub Issues for automatic improvement tracking
• Support for multiple project types beyond Microsoft Store
• Machine learning on captured patterns across projects
• Team collaboration features for shared learning
• Automated pattern detection from CI/CD logs

Contributing

The Weaver system is designed to evolve based on captured lessons. Contributions welcome:

1. Share your evolution logs (sanitized)
2. Contribute new investigation patterns
3. Improve question generation algorithms
4. Add new professional perspectives

License

MIT

Acknowledgments

• BMAD Method for enforcement patterns
• The "zealous liar" pattern that started it all
• Every failed build that taught us something