# PM Agent MCP Server

## Overview
MCP server with 4 tools to help Product Manager (Asha) make data-driven decisions.

## The 4 Tools

### 1. prioritize_backlog
Ranks backlog items (1-35) by RICE scoring.
- **Input:** method, max_results, filters
- **Output:** Ranked items with flags (dependencies, stale, unestimated, no customer signal)

### 2. analyze_feedback
Extracts themes from 90 customer feedback entries.
- **Input:** group_by, sentiment_filter, bias_analysis flag
- **Output:** Themes with customer segments, bias warnings

### 3. assess_capacity
Calculates real team capacity for a sprint.
- **Input:** sprint_id, engineer names (optional)
- **Output:** Team capacity + per-engineer breakdown with warnings
- **Formula Discovered:** `available = (21 - pto_days × 2.1) × (allocation/100) - carry_over`

### 4. map_dependencies
Maps dependency chains for backlog items.
- **Input:** item_ids, max_depth
- **Output:** Dependency graph, cycles detected, risk flags

## Setup

```bash
python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
python server.py
```

## Data Path Contract

The server reads data from `PM_AGENT_DATA` environment variable:
```bash
export PM_AGENT_DATA=/path/to/data
python server.py
```

Falls back to `./data` if env var not set.

## Tools Usage

Each tool returns JSON with:
- `status`: "success" or "error"
- `data`: Tool-specific output
- `message`: Error details if status is "error"

## Files

- `server.py` - MCP server entry point
- `tools/` - Tool implementations
- `data/` - Sample data for local testing
- `requirements.txt` - Dependencies