The Multi Agent Orchestration Master Library is a skill — not an MCP server — published by agensi. It provides a collection of patterns designed to help Claude Code manage and coordinate multiple AI agents running in parallel. The library covers orchestration through tmux sessions, YAML-based task queues, and event-driven communication protocols. It includes hierarchical command structures that define how a directing agent delegates work, handoff documents that transfer context between agents at task boundaries, and parallel dispatch patterns for splitting work across concurrent worker agents. Compaction-recovery procedures address what happens when an agent hits context limits mid-task, allowing the orchestration to resume without losing progress. The library is described as a meta-skill: it does not perform domain-specific work itself, but instead extends Claude Code’s ability to spawn, direct, and coordinate subordinate agents. Developers working on large, multi-step coding tasks that benefit from parallel execution — such as running independent sub-tasks simultaneously — are the primary audience. Because this is a skill rather than an MCP server with discrete tools, it does not expose callable functions to an agent runtime in the conventional sense; it provides reusable orchestration patterns and procedures that the host agent applies.

Use cases

Coordinate multiple Claude Code worker agents across parallel sub-tasks using tmux sessions

Manage task handoffs between agents using structured handoff documents

Define hierarchical command structures so a directing agent can delegate to workers

Use YAML task queues to sequence and dispatch work across concurrent agents

Recover agent orchestration state after a context compaction event

Apply event-driven protocols to synchronize progress across spawned agents

When to use it

When a single Claude Code session is insufficient to complete a large, parallelisable coding task

When tasks can be decomposed into independent units suitable for separate worker agents

When context-limit recovery is needed across long-running multi-agent workflows

When a tmux-based local environment is available to host multiple agent sessions

When not to use it

When looking for an MCP server with callable tools — this skill exposes no tool endpoints

When the runtime does not support tmux or cannot spawn multiple agent sessions

When the task is self-contained and does not benefit from parallel agent coordination

When a non-Claude Code agent framework is in use, as patterns are specific to Claude Code