Claude-Flow Integration

Parallel agent execution for faster development.

What is Claude-Flow?

An MCP server enabling multi-agent coordination:

• Swarm topologies
• Parallel agent spawning
• Task orchestration
• Memory persistence

Prerequisites

Install claude-flow:

npm install -g claude-flow@alpha

Configure MCP server in Claude Code settings.

Topologies

Hierarchical

Tree structure with coordinator.

swarm_init({
  topology: "hierarchical",
  strategy: "balanced",
  maxAgents: 8
})

Use for: Coordinated tasks, phase-based work

Mesh

Peer-to-peer, no central coordinator.

swarm_init({
  topology: "mesh",
  strategy: "balanced",
  maxAgents: 8
})

Use for: Independent parallel research

Star

Central hub with worker spokes.

swarm_init({
  topology: "star",
  strategy: "balanced",
  maxAgents: 8
})

Use for: Aggregation tasks, result collection

Ring

Circular message passing.

swarm_init({
  topology: "ring",
  strategy: "balanced",
  maxAgents: 8
})

Use for: Pipeline processing

Parallel Agent Spawning

Sequential spawning is slow. Use batch:

agents_spawn_parallel({
  agents: [
    {type: "researcher", name: "research-oauth"},
    {type: "researcher", name: "research-jwt"},
    {type: "researcher", name: "research-security"}
  ],
  maxConcurrency: 5,
  batchSize: 3
})

Result: 10-20x faster than sequential.

Task Orchestration

task_orchestrate({
  task: "Research authentication patterns",
  strategy: "parallel",  // or sequential, adaptive
  priority: "high",
  maxAgents: 3
})

Strategies:

• parallel - All at once
• sequential - One after another
• adaptive - System decides based on load

Patterns

Parallel Research

When /plan:hard needs multiple research threads:

1. Init mesh topology
2. Spawn researchers in parallel
3. Each researches independently
4. Aggregate results

swarm_init({topology: "mesh"})

agents_spawn_parallel({
  agents: [
    {type: "researcher", name: "api-research"},
    {type: "researcher", name: "security-research"},
    {type: "researcher", name: "performance-research"}
  ]
})

task_orchestrate({
  task: "Research OAuth implementation",
  strategy: "parallel"
})

Multi-Phase Implementation

For /code:parallel:

1. Init star topology
2. Coordinator manages phases
3. Workers implement in parallel
4. Merge changes

swarm_init({topology: "star"})

// Phase with independent tasks
agents_spawn_parallel({
  agents: [
    {type: "coder", name: "impl-models"},
    {type: "coder", name: "impl-routes"},
    {type: "coder", name: "impl-views"}
  ]
})

Test Parallelization

swarm_init({topology: "star"})

agents_spawn_parallel({
  agents: [
    {type: "tester", name: "test-unit"},
    {type: "tester", name: "test-integration"},
    {type: "tester", name: "test-e2e"}
  ]
})

task_orchestrate({
  task: "Run all test suites",
  strategy: "parallel"
})

Performance Reference

Operation	Sequential	Parallel	Speedup
Agent spawn (5)	5s	0.3s	16x
Research (3 topics)	30s	12s	2.5x
Test suites (4)	60s	18s	3.3x

MCP Tools Reference

Tool	Purpose
swarm_init	Create topology
agents_spawn_parallel	Batch spawn
agent_spawn	Single agent
task_orchestrate	Run task
swarm_status	Monitor health
agent_list	List agents
agent_metrics	Performance
memory_usage	Persistence

Related

• SPARC Methodology - Agent roles
• Coding Commands - /code:parallel