Integration Guides

Integration Guides

MCP Task Orchestrator can be integrated at different levels of sophistication. These guides are progressive — each tier builds on the previous. Start where you are, level up when you need more.

Guide Overview

Tier	Guide	Audience	What You Get
1	Bare MCP	Any MCP client user	13 tools, persistent SQLite, role-based workflow
2	CLAUDE.md-Driven	Claude Code users	Consistent agent behavior via project instructions
3	Note Schemas	Schema configurers	Phase gate enforcement, required documentation
4	Plugin: Skills and Hooks	Claude Code + plugin	Automated workflows, plan-mode pipeline, subagent protocols
5	Output Styles	Power users	Full orchestrator mode with delegation
6	Self-Improving Workflow	Self-optimizers	Feedback loop, observation logging, auto-memory correction
7	Ralph Loop	Backlog drainers, fleet operators	Autonomous queue drain — fresh-context-per-iteration, claim mode, schema-driven

Which Guide Is Right for Me?

• "I use a non-Claude-Code MCP client" → Tier 1
• "I use Claude Code but don't want a plugin" → Tiers 1-2
• "I want to enforce documentation quality at each phase" → Add Tier 3 to any tier
• "I want automated skill workflows and plan-mode integration" → Tier 4
• "I want Claude to orchestrate and delegate, not implement directly" → Tier 5
• "I want the system to improve itself over time" → Tier 6
• "I want to drain a backlog of independent items autonomously" → Tier 7

Layering

Each tier adds on top of the previous. Note schemas (Tier 3) can be added independently at any tier — they require only a .taskorchestrator/config.yaml file and no plugin or CLAUDE.md changes. The self-improving workflow (Tier 6) is also relatively standalone — it can be adapted with or without the full output style.

The Ralph Loop (Tier 7) is an alternate execution mode rather than a strict superset. It builds on Tier 4 (plugin) but operates differently from Tier 5: where Tier 5 is a single-session orchestrator that plans and delegates, Tier 7 spawns a fresh claude -p per item and walks the backlog autonomously. Pair it with Tier 3 schemas to drive what each iteration does, or with Tier 6 to capture cross-session signal from drain runs.