An MCP (Model Context Protocol) server that exposes product-management tools for AI agents. Built with the [MCP Python SDK](https://github.com/modelcontextprotocol/python-sdk) using `FastMCP`.

## Tools

| Tool | Description |
|------|-------------|
| `prioritize_backlog` | Rank backlog items using RICE, value/effort, or customer-signal scoring |
| `analyze_feedback` | Extract and rank themes from customer feedback |
| `assess_capacity` | Calculate per-engineer sprint capacity with carry-over and skill-fit checks |
| `map_dependencies` | Trace dependency chains via BFS and surface risks |

## Project Structure

```
ProductNerveCenter/
├── server.py                      # MCP server — data loading + tool wrappers
├── olympics.json                  # Evaluation agent configuration
├── tools/
│   ├── __init__.py                # Package exports
│   ├── prioritize_backlog.py      # RICE / value-effort / customer-signal scoring
│   ├── analyze_feedback.py        # Theme extraction & grouping logic
│   ├── assess_capacity.py         # Sprint capacity calculation
│   └── map_dependencies.py        # BFS dep traversal & risk analysis
├── data/
│   ├── product_backlog.json       # 35 backlog items
│   ├── customer_feedback.json     # 90 customer feedback entries
│   ├── team_roster.json           # 8 engineers across 2 squads
│   ├── dependencies.json          # Dependency graph edges
│   └── sprint_history.json        # 6 sprint history records
├── oracle_connection/
│   └── README.md                  # Discovery process documentation
├── TECHNICAL_DECISIONS.md         # Design decisions log
├── data_dictionary.md             # Field definitions for all data files
├── requirements.txt
├── agent_config.json
└── env_vars.json
```

## Prerequisites

- **Python 3.11 or 3.12**
- pip

## Setup

```bash
# Set Python version (if using pyenv)
pyenv local 3.11.11  # or 3.12.3

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

# Install dependencies
pip install -r requirements.txt
```

## Running the Server

### stdio mode (for Claude Desktop / evaluation agent)

```bash
python3 server.py
```

### HTTP mode (for browser/network clients)

```bash
python3 server.py http 8000
```

This starts a Streamable HTTP server at `http://127.0.0.1:8000`.

## Connecting to the Server

### From Claude Desktop

Add to your `claude_desktop_config.json`:

```json
{
  "mcpServers": {
    "devpulse": {
      "command": "python3",
      "args": ["/full/path/to/ProductNerveCenter/server.py"],
      "env": {
        "PM_AGENT_DATA": "/full/path/to/ProductNerveCenter/data",
        "MCP_DATA_URL": "https://co-mcp-server-dev.apps-internal.lrl.lilly.com/mcp"
      }
    }
  }
}
```

### From another MCP client (HTTP mode)

```python
from mcp.client.streamable_http import streamablehttp_client
from mcp import ClientSession

async with streamablehttp_client("http://127.0.0.1:8000/mcp") as (read, write, _):
    async with ClientSession(read, write) as session:
        await session.initialize()
        result = await session.call_tool("prioritize_backlog", {"method": "rice"})
        print(result)
```

## Environment Variables

| Variable | Purpose | Default |
|----------|---------|---------|
| `PM_AGENT_DATA` | Path to the `data/` folder with JSON files | `./data` |
| `MCP_DATA_URL` | Remote MCP server URL for team roster & dependency map | `https://co-mcp-server-dev.apps-internal.lrl.lilly.com/mcp` |

## Quick Test (no remote server needed)

The server gracefully handles the remote MCP server being unreachable — it logs a warning and continues with empty roster/deps. You can run it locally right away:

```bash
source .venv/bin/activate
python3 server.py http 8000
```

You'll see:

```
[MCP] ... WARNING MCP server unreachable (...) — roster and deps will be empty
```

The 4 tools will still work using the local JSON files in `data/`.

## Tool Details

### `prioritize_backlog`

| Parameter | Type | Default | Description |
|-----------|------|---------|-------------|
| `method` | string | `"value_effort"` | Scoring method: `"rice"`, `"value_effort"`, `"customer_signal"` |
| `filters` | dict | `null` | Filter by `squad`, `status`, or `tags` |
| `include_dependency_check` | bool | `true` | Flag items with unresolved blockers |

### `analyze_feedback`

| Parameter | Type | Default | Description |
|-----------|------|---------|-------------|
| `time_range` | dict | `null` | `{"start": "YYYY-MM-DD", "end": "YYYY-MM-DD"}` |
| `customer_tier` | string | `null` | `"enterprise"`, `"mid_market"`, or `"startup"` |
| `source` | string | `null` | `"support_ticket"`, `"nps_survey"`, `"sales_call"`, `"user_interview"` |
| `group_by` | string | `"theme"` | `"theme"`, `"customer"`, or `"source"` |

### `assess_capacity`

| Parameter | Type | Default | Description |
|-----------|------|---------|-------------|
| `sprint_id` | string | `null` | Target sprint ID (uses latest if null) |
| `squad` | string | `"all"` | Filter by squad name |
| `include_carry_over` | bool | `true` | Subtract in-progress points from capacity |
| `check_skill_fit` | bool | `false` | Flag skill mismatches on assigned items |

### `map_dependencies`

| Parameter | Type | Default | Description |
|-----------|------|---------|-------------|
| `item_ids` | list | `null` | Item IDs to trace (null = all in-progress/planned) |
| `include_external` | bool | `true` | Include external dependencies |
| `max_depth` | int | `3` | Maximum BFS traversal depth |