Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/ruvnet/ruflo/llms.txt

Use this file to discover all available pages before exploring further.

This quickstart walks you from zero to a fully working Ruflo setup with Claude Code in under 5 minutes. By the end you will have Ruflo initialized in your project, the MCP server registered so Claude Code can call all 313 tools, and a verified installation ready to spawn agent swarms on demand.
1

Prerequisites

Ruflo requires Node.js ≥ 20.0.0. Check your current version:
node --version
# Should print v20.0.0 or higher
If your version is below 20, install the latest LTS from nodejs.org or use a version manager like nvm:
nvm install 20
nvm use 20
Claude Code must also be installed. If you haven’t installed it yet:
npm install -g @anthropic-ai/claude-code
2

Install Ruflo

Choose your install method. The interactive wizard is recommended for first-time setup because it walks through configuration options and validates the environment.
# Recommended for first-time setup — interactive prompts guide configuration
npx ruflo@latest init wizard
The init command creates .claude/, .claude-flow/, CLAUDE.md, hook handlers, and settings in your project. Nothing is written outside your project directory.
Windows users: The curl ... | bash one-liner requires a POSIX shell (Git-Bash, WSL, or MSYS). If you see 'bash' is not recognized, use npx ruflo@latest init wizard instead — it runs identically in native PowerShell and produces the same result.
3

Add the MCP Server to Claude Code

Register Ruflo as an MCP server so Claude Code can call all 313 Ruflo tools directly from your sessions:
claude mcp add ruflo -- npx ruflo@latest mcp start
Verify the server is registered:
claude mcp list
# ruflo   npx ruflo@latest mcp start
Once registered, tools like swarm_init, agent_spawn, memory_store, memory_search, and hooks_route are available to Claude Code in every session without any further configuration.
4

Verify Installation

Confirm Ruflo is installed and healthy:
# Check the installed version
npx ruflo@latest --version

# Run the built-in diagnostics (checks Node version, MCP server, hooks, memory)
npx ruflo@latest doctor
The doctor command validates your environment and prints a checklist. All items should show ✅. If any show ❌, the output includes a remediation hint.
5

Launch Your First Swarm

Use the hive-mind spawn command to launch a queen-led agent swarm with a plain-English objective:
npx ruflo@latest hive-mind spawn "Implement user authentication"
What happens next:
  1. A Strategic Queen agent decomposes the objective into a task plan
  2. The queen spawns specialized workers — architect, coder, tester, reviewer — from the 100+ available agent types
  3. Workers claim tasks, collaborate via shared memory, and produce output in parallel
  4. The post-task hook fires on completion, storing successful patterns in HNSW vector memory for future routing
You can also spawn a swarm with explicit queen type and consensus algorithm:
npx ruflo@latest hive-mind spawn "Refactor auth module for OAuth 2.0" \
  --queen-type tactical \
  --consensus raft
Check swarm status at any time:
npx ruflo@latest hive-mind status

​
Basic Usage Examples

# Interactive setup wizard
npx ruflo@latest init wizard

# Non-interactive with defaults
npx ruflo@latest init

# Initialize for Codex CLI instead of Claude Code
npx ruflo@latest init --codex

# Initialize for both Claude Code and Codex (dual mode)
npx ruflo@latest init --dual

​
The Learning Loop

After each task completes, the hooks system stores successful patterns in HNSW-indexed vector memory. The next time you run a similar task, the router retrieves those patterns (sub-millisecond search) and routes work to the agents that performed best — no manual configuration required. The system gets measurably smarter the more you use it.
The learning loop runs automatically:
  1. Pre-task hook fires → router searches memory for similar past tasks
  2. Agents execute the task using retrieved context
  3. Post-task hook fires → successful patterns are stored with embeddings
  4. Router updates → future similar tasks hit the best-performing path
You can inspect learning state at any time: 
npx ruflo@latest hooks intelligence --status
npx ruflo@latest neural status
For deeper Claude Code integration — including how hooks interact with Claude’s session lifecycle, how to configure tool groups to reduce latency, and how to use the CLAUDE_FLOW_TOOL_MODE preset — see the Claude Code Integration guide.

Build docs developers (and LLMs) love

Get started for freeTalk to us