MAW-MCP: Multi-Agent Workflow MCP Server

Coordinate parallel AI agent development workflows from any Claude interface.

What It Does

• Review: Analyze codebase, identify improvements, generate agent prompts
• Launch: Get copy-paste prompts with branch names and sequencing
• Integrate: Merge order + comprehensive test plan
• Decide: Deploy, iterate, or add features recommendation

Installation

# Clone
git clone git@github.com:derekparent/maw-mcp.git
cd maw-mcp

# Create virtual environment
python3 -m venv venv
source venv/bin/activate

# Install
pip install -e .

Configuration

Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "maw": {
      "command": "python",
      "args": ["-m", "src.server"],
      "cwd": "/Users/dp/Projects/maw-mcp"
    }
  }
}

Cursor

Add to ~/.cursor/mcp.json:

{
  "mcpServers": {
    "maw": {
      "command": "/Users/dp/Projects/maw-mcp/venv/bin/python",
      "args": ["-m", "src.server"],
      "cwd": "/Users/dp/Projects/maw-mcp"
    }
  }
}

Tools

Tool	Description
maw_status	Show current workflow state and next action
maw_review	Analyze codebase, generate agent prompts
maw_launch	Get agent prompts with sequencing. Use next=true for one-at-a-time mode
maw_checkin	Aggregate progress reports, show dashboard
maw_integrate	Merge order + test plan
maw_decide	Deploy/iterate/add recommendation
maw_learn	Capture learnings to PROJECT_LEARNINGS.md
maw_patterns	Search accumulated patterns

Workflow

idle → review → launch → integrate → decide
          ↑                             ↓
          └─────────────────────────────┘
                    (iterate)

Review

> maw_review focus="security" wave_name="Security Hardening"

[Claude analyzes codebase and creates AGENT_PROMPTS/]

Review AGENT_PROMPTS/ before running maw_launch

Focus Options

Use shortcuts or natural language:

Shortcut	Also matches
ui	ux, design, styling, layout, responsive, forms, buttons, mobile
api	endpoints, routes, rest, graphql
data	database, db, schema, queries, sql, models, migrations
performance	speed, slow, fast, optimize, caching, memory
testing	tests, coverage, tdd, pytest, e2e
security	auth, secrets, passwords, injection, xss
docs	documentation, readme, comments, docstrings
architecture	structure, refactor, modules, organization
errors	error handling, exceptions, logging, debugging
dx	setup, tooling, config, developer experience

Or just say what you want:

> maw_review focus="make the forms less janky"
> maw_review focus="error handling is weak"
> maw_review focus="needs better offline support"

GOALS.md

Create GOALS.md in your repo to steer MAW automatically:

# Project Goals

## Active Focus
- Mobile-first redesign of inventory screens
- Touch targets too small

## Backlog
- Add offline sync
- Better error messages

## Not Now
- Performance optimization
- Dark mode

MAW reads this and prioritizes findings accordingly. The --focus flag overrides GOALS.md when provided.

Launch

> maw_launch

## 🚀 Agent Launch Sequence

### Launch Order
1. Agent 1 - Run first (~20 min)
2. Agents 2, 3, 4 - Run in parallel

### Agent Prompts

#### Agent 1: Backend Security
**Branch:** `agent/1-backend-security`
...

Sequential Mode

For smaller projects or when you can't run multiple agents, use sequential mode:

> maw_launch next=true

## 🔢 Task 1 of 4

### Agent 1: Backend Security
**Branch:** `agent/1-backend-security`
...

*Run `maw_launch next=true` again for the next task (3 remaining)*

Call maw_launch next=true repeatedly to work through tasks one at a time. State persists across sessions.

Check In

Agents provide completion reports when done:

> maw_checkin reports="## Agent 1 Completion Report..."

## 📊 Agent Status Dashboard

| Status | Count |
|--------|-------|
| ✅ Complete | 3 |
| ⚠️ Partial | 1 |

Integrate

> maw_integrate

## Integrate

### Recommended Merge Order
1. Documentation (lowest risk)
2. Tests
3. Backend
4. Frontend

## 🧪 Pre-Integration Test Plan

| Test | Expected | Status |
|------|----------|--------|
| App starts | Clean startup | ☐ |
| GET /api/health | 200 OK | ☐ |
...

Decide

> maw_decide

## Decide: What's Next?

| Condition | Recommendation |
|-----------|----------------|
| All tests pass | ✅ Deploy |
| Quality < 7/10 | 🔄 Iterate |

What Gets Created in Projects

your-project/
├── WORKFLOW_STATE.json     # Tracks phase, iteration, agents
├── AGENT_PROMPTS/          # Generated by maw_review
│   ├── 1_Backend_Engineer.md
│   ├── 2_Security.md
│   ├── COORDINATION.md
│   └── README.md
└── PROJECT_LEARNINGS.md    # Session notes

Philosophy

• Soft enforcement: Warnings, not hard blocks
• You're the operator: Tools inform, you decide
• Deliberate pacing: Review before launch, check before integrate
• Minimal project footprint: Guides stay in server, not copied everywhere

Content

Guides and templates in content/:

content/
├── phases/
│   ├── review.md
│   ├── launch.md
│   ├── integrate.md
│   └── decide.md
├── templates/
│   ├── AGENT_PROMPT.md
│   ├── COORDINATION.md
│   ├── GOALS.md          # Copy to your repo
│   └── PR_TEMPLATE.md
└── UNIVERSAL_PATTERNS.md

License

MIT