claude-flow

name	claude-flow
description	Orchestrate multi-agent swarms and tasks via claude-flow v3 MCP
version	1.0.0
allowed-tools	["claude-flow__swarm_*","claude-flow__agent_*","claude-flow__task_*","claude-flow__memory_*","claude-flow__session_*","claude-flow__hive-mind_*","claude-flow__workflow_*","claude-flow__coordination_*","claude-flow__performance_*","claude-flow__system_*"]
user-invocable	true
argument-hint	Swarm objective or task description

Claude Flow v3 MCP Skill

This skill provides access to claude-flow v3 swarm orchestration via the claude-flow MCP server. It enables multi-agent coordination, task management, shared memory, and workflow automation.

Key Capability Categories

Swarm Management (swarm_*)

Initialize, monitor, and shut down multi-agent swarms. Supports hierarchical, mesh, and adaptive topologies with configurable agent counts and strategies.

Agent Lifecycle (agent_*)

Spawn, list, monitor, update, and terminate individual agents. Supports 60+ agent types including coders, reviewers, testers, planners, and specialized roles like security-architect and performance-engineer.

Task Orchestration (task_*)

Create, list, update, complete, and cancel tasks. Tasks can be assigned to agents and tracked through their lifecycle with status updates.

Shared Memory (memory_*)

Store, retrieve, search, list, and delete shared knowledge across agents. Supports namespaces, TTL, tags, and HNSW-powered semantic search for coordination between agents working on related problems.

Session Persistence (session_*)

Save, restore, list, and manage session state. Enables checkpointing of agent progress and resuming work across restarts.

Hive-Mind Consensus (hive-mind_*)

Initialize hive-mind clusters with Byzantine fault-tolerant consensus (Raft). Broadcast messages, reach consensus on decisions, manage shared memory, and coordinate agent groups for collective decision-making.

Workflow Automation (workflow_*)

Create, execute, pause, resume, cancel, and delete workflows. Supports workflow templates for repeatable multi-step processes with agent orchestration.

Coordination (coordination_*)

Node registration, topology management, load balancing, synchronization, consensus, and metrics collection for distributed agent coordination.

Performance Monitoring (performance_*)

Run benchmarks, detect bottlenecks, collect metrics, generate profiles and reports, and apply optimizations across the agent infrastructure.

System Health (system_*)

Check system status, health, metrics, and info. Reset system state when needed.

Common Usage Patterns

• Spawning multi-agent swarms: Use swarm_init to set up topology, then agent_spawn to create specialized agents for parallel work.
• Coordinating parallel tasks: Use task_create to define work items, assign them to agents, and track completion with task_status.
• Sharing knowledge across agents: Use memory_store and memory_search to let agents share findings, patterns, and decisions.
• Running workflows: Use workflow_create with templates, then workflow_execute for repeatable multi-step processes.
• Consensus decisions: Use hive-mind_init and hive-mind_consensus for collective decision-making across agent groups.

Configuration Note

The claude-flow MCP server is configured as internal_only, meaning it is not exposed to external consumers. Tools are namespaced as claude-flow__<tool> (e.g. claude-flow__swarm_init, claude-flow__memory_store).

Reference

• Documentation: https://github.com/ruvnet/claude-flow/tree/main/v3