bobbyrgoldsmith/quarterback

**Read the field. Call the play.**
Strategic task prioritization and agent orchestration for multi-project operators.
---
Every other AI task manager breaks down **one project** into subtasks. Quarterback helps you decide which of your **ten projects** to prioritize right now — using a 5-factor weighted scoring engine, organizational context, and time-aware planning. It runs locally, costs nothing, and works as both a standalone CLI and an MCP server for Claude.
What Makes Quarterback Different
| Feature | Quarterback | TaskMaster AI | Shrimp Task Manager |
|---------|-------------|---------------|---------------------|
| **Multi-project prioritization** | 5-factor weighted engine | Single-project breakdown | Single-project |
| **Advisory document system** | Analyze articles against your goals | No | No |
| **Agent orchestration** | Autonomy levels + webhooks | No | No |
| **Time-aware planning** | Working hours, lunch, buffer time | No | No |
| **Organizational context** | Goals, constraints, workflows | No | No |
| **Knowledge wiki (Playbook)** | LLM-maintained wiki for cross-session consistency | No | No |
| **Conflict detection** | Cross-project scheduling conflicts | No | No |
| **Standalone CLI** | Full CLI without AI runtime | Requires AI | Requires AI |
| **Cost** | Free (MIT) | Free | Free |
Quick Start
LLM-Powered Setup (via MCP)
When using Quarterback as an MCP server, ask your LLM: *"Set up Quarterback for me"* — it will call the setup_quarterback tool, interview you conversationally about your business, goals, workflows, projects, constraints, and knowledge base (Playbook), then write all config files and database records in one shot. No manual YAML editing required.
MCP Server
Quarterback works with **any MCP-compatible client** — Claude Desktop, Claude Code, Cursor, Windsurf, Cline, OpenAI agents, and others. All 23 tools use standard MCP protocol (JSON-RPC over stdio) with no LLM-specific dependencies.
Add to your Claude Desktop config (~/Library/Application Support/Claude/claude_desktop_config.json):
Or for Claude Code (~/.claude/settings.json):
The same quarterback-server command works with any MCP client — just add it to your client's server configuration.
Then ask your LLM: *"What should I work on today?"* — it will use all 23 Quarterback tools to analyze your priorities.
Features
5-Factor Prioritization Engine
Every task is scored across five dimensions:
| Factor | Weight | What it measures |
|--------|--------|-----------------|
| **Impact** | 30% | Task impact + project revenue/strategic value |
| **Urgency** | 25% | Due date proximity + blocking status |
| **Strategic** | 25% | Project priority + milestone status |
| **Effort** | 15% | Inverted effort score (quick tasks score higher) |
| **Quick Win** | 5% | High impact + low effort bonus |
Advisory Document System
Analyze external articles, books, and advice against your organizational context:
The analyzer checks every recommendation against your goals and constraints, flagging conflicts and synergies.
Agent Orchestration
Mark tasks for autonomous agent execution with configurable autonomy:
- **Draft**: Agent creates a draft for your review
- **Checkpoint**: Agent pauses at key decisions for approval
- **Autonomous**: Agent runs to completion
Webhooks notify your automation layer (n8n, Zapier, custom) when tasks are ready.
Playbook — Knowledge Wiki
Playbook is Quarterback's compiled knowledge layer. It's an LLM-maintained markdown wiki that gives every session — local CLI, MCP, or autonomous agent — the same canonical context about your projects, decisions, and strategies.
**The problem it solves:** Without Playbook, each AI session starts fresh and re-derives your organizational context from sparse signals. Two sessions running the same query can produce different results because they reconstruct understanding independently. Playbook provides accumulated knowledge that all sessions read from.