AI orchestration with hive-mind swarms, neural networks, and 87 MCP tools for enterprise dev.
io.github.ruvnet/claude-flow MCP Server
This MCP server is described as an AI orchestration solution using “hive-mind swarms” and neural networks, and it provides 87 MCP tools for enterprise development workflows. The server’s identity is listed as io.github.ruvnet/claude-flow, with focus areas that include agentic systems and multi-agent coordination.
🛠️ Key Features
AI orchestration using hive-mind swarms
Neural networks
87 MCP tools
Topics include multi-agent, swarm intelligence, and autonomous agents
🚀 Use Cases
Enterprise development workflows that require MCP tool access
Multi-agent systems and agentic workflows
Agentic coordination across “skills” and multi-agent setups
⚡ Developer Benefits
Provides a large set of MCP tools (87)
Integrates with ecosystems referenced by topics (e.g., npm, TypeScript)
Supports “agentic workflow” and multi-agent system development concepts
⚠️ Limitations
Source data does not specify tool names, configuration, authentication, or supported platforms beyond described topics
Then open Claude Code in the same folder and give it a real task, for example:
text
Use a ruflo swarm to review this repo: find the riskiest module, write tests for it, and summarise what you changed.
init adds ruflo's agents, hooks, memory and MCP server to the project, and Claude Code picks them up on its next start. Codex and other MCP clients work too: point them at npx -y ruflo@latest mcp start.
Check the result. Ask the agent to name the module it chose, show the test it added, and report the test command and result. Review the diff in your project.
Looping recording of the ruflo console: overview, swarm view and missions in a sample project
Only if ruflo-core is installed (it ships its own .mcp.json) — most other plugins don't
Yes
Hooks installed
No
Yes
Best for
Try a single plugin's commands without committing to the full install
Production use — everything works as documented
Claude Code plugins only. This route installs core tools, the visual console and runtime mods, with no files added to your project. It needs Claude Code 2.1.287 or later. Run inside Claude Code:
From a terminal, on Claude Code 2.1.292 or later: claude plugin install ruflo-core@ruflo --marketplace ruvnet/ruflo. Mods run with your account's permissions, so review their code before installing.
Codex or another MCP client, without project setup. Configure the commandnpx with the arguments["-y", "ruflo@latest", "mcp", "start"]. To set up a project interactively, run npx ruflo@latest init wizard.
Plain terminal. The console also runs on its own, without Claude Code: npx ruflo tui. Add --plain to print one frame for scripts. See the terminal UI guide.
ruflo-core registers its own MCP server. Its tools use the mcp__plugin_ruflo-core_ruflo__ prefix.
CLI Install
macOS / Linux / WSL / Git-Bash:
bash
# One-line install (POSIX shells only — see Windows note below)
curl -fsSL https://cdn.jsdelivr.net/gh/ruvnet/ruflo@main/scripts/install.sh | bash
All platforms (including native Windows PowerShell / cmd):
bash
# Interactive setup wizard — runs identically on every platform
npx ruflo@latest init wizard
# Quick non-interactive init# npx ruflo@latest init# Or install globally
npm install -g ruflo@latest
💡 Windows users: the curl ... | bash form needs a POSIX shell (Git-Bash, WSL, MSYS). The npx ruflo@latest init wizard line works natively in PowerShell and cmd. If you hit an 'bash' is not recognized error, use the npx line instead — both end up running the same init flow.
MCP Server
bash
# Add Ruflo as an MCP server in Claude Code
claude mcp add claude-flow -- npx ruflo@latest mcp start
For static inspection of local applications and binaries, see the ruRE plugin guide.
Grok Build:npx ruflo init --grok writes the project host surface (.grok/, the team bus, and the operator guide) and nothing outside the project; add --dry-run to preview. Setup, the spawn_subagent contract, the optional status row, and what to re-check when Grok updates: the Grok operator guide.
Mission Control
Open /ruflo in Claude Code to monitor agent work, inspect a run, or give Claude permission to operate the console.
If you use the console, open the agent’s log and compare its actions with the diff and test result from your first task.
Open /plugin to check the active mods, then /ruflo to launch the console. For a project initialized through the CLI, run npx ruflo init and restart Claude Code before opening /ruflo.
Console mods and catalog
Mod
Purpose
ruflo-console
Opens the console beside Claude Code.
ruflo-mods
Routes prompts and enforces policy in process.
ruflo-swarm
Adds a pane showing the swarm.
ruflo-ruos
Adds the ruOS status segment.
Open /ruflo market to inspect plugins, mods, and skills. The catalog asks for confirmation before installing, enabling, disabling, or updating an item.
Create a mission with Claude
The default is read + ask. Available levels are read, write, manage, and full, with ask or auto approval. Actions above the selected level are refused. Network, spending, and destructive actions require confirmation even in auto.
In Settings → Claude control, select write + ask.
Ask Claude, “Create a mission to add a dark mode toggle. Show me the mission and its status.”
Confirm the pending creation request. Check Missions for the goal and Overview for the action log. This verifies that the mission was created.
Select Take back control, then ask Claude to take another console action. The console should refuse it until you restore access.
Specialized agents for coding, testing, security, docs, architecture
Comms Layer
Zero-trust federation — agents across machines/orgs discover, authenticate, and exchange work securely
Swarm Coordination
Hierarchical, mesh, and adaptive topologies with consensus
Self-Learning
SONA neural patterns, ReasoningBank, trajectory learning
Vector Memory
HNSW-indexed AgentDB — measured ~1.9x faster at N=20k, ~3.2x–4.7x at N=5k vs brute force (recall@10 ~0.99); ANN wins above the crossover, ties/loses at small N. See audit + scripts/benchmark-intelligence.mjs
Audit your AI agent setup before you ship. Grade readiness (1-100), scan tool configs for security issues, snapshot the whole project to catch regressions over time, and find templates that match your repo. ruflo eject turns a ruflo project into a standalone agent toolkit with its own name. Full guide.
External, maintained by Agentic QE: 11 QE agents, test generation, coverage, chaos and TDD skills, the agentic-qe MCP server, and a guard for AQE learning data (/plugin install agentic-qe-fleet@ruflo). Install it from one marketplace only, agentic-qe or ruflo: enabled from both, its mod and commands load twice
External, maintained in ruvnet/agentdb: agentdb's own MCP server (pinned agentdb@3.2.0: hybrid keyword + vector search, opt-in cross-encoder rerank), a memory-specialist agent, vector-memory / learning / diagnostics skills, /agentdb-status, and a guard that keeps secrets out of memory and refuses delete/reset unless allowed (/plugin install agentdb@ruflo). Use ruflo-agentdb instead if you want AgentDB through ruflo's server; the two can coexist. Install it from one marketplace only, agentdb or ruflo
External, maintained in ruvnet/agentic-flow: the agentic-flow MCP server (pinned agentic-flow@3.1.0, readonly profile by default, nothing pre-approved), ReasoningBank and model-router agents, routing / learning / autopilot skills, /agentic-flow-status, and a guard on agentic_flow_agent (/plugin install agentic-flow@ruflo). Install it from one marketplace only, agentic-flow or ruflo
Run agents — local WASM sandbox (rvagent) + Anthropic Claude Managed Agents (cloud), with opt-in bounded dynamic workflows, gated session controls and verified advisory learning
Compose and generate music through a Cognitum Music account
How Ruflo works
Agent = Model + Harness. The model writes; the harness gives it tools, memory, loops, sandboxes, and controls so it can actually work. Ruflo is the harness — the execution layer around Claude Code and Codex that adds 100+ specialized agents, coordinated swarms, self-learning memory, federated comms across machines, and enterprise security guardrails. So agents don't just run, they collaborate.
What can the next run reuse?
Useful trajectories and task feedback can inform future work. Results depend on the configured memory, learning components and quality of feedback. Learning plugin.
Slack gave teams channels. Federation gives agents the same thing — shared workspaces across trust boundaries, where agents on different machines, orgs, or cloud regions can discover each other, prove who they are, and collaborate on tasks.
The difference: some channels are trusted, some aren't. @claude-flow/plugin-agent-federation handles that automatically. Your agents join a federation, get verified via mTLS + ed25519, and start exchanging work — with PII stripped before anything leaves your node and every message auditable. Untrusted agents can still participate at lower privilege: they see discovery info, not your memory. As they prove reliable, trust upgrades. If they misbehave, they get downgraded instantly — no human in the loop required.
You don't configure handshakes or manage certificates. You federation init, federation join, and your agents start talking. The protocol handles identity, the PII pipeline handles data safety, and the audit trail handles compliance.
📘 Full user guide:docs/federation/ — setup, MCP tools, trust levels, circuit breaker, and the (opt-in) WireGuard mesh layer that ties packet-layer reachability to federation trust. ADR-111 deep-dive at docs/federation/phase7-mesh-bringup.md.
Federation capabilities
Capability
How it works
🔒
Zero-trust federation
Remote agents start untrusted. Identity proven via mTLS + ed25519 challenge-response. No API keys, no shared secrets.
🛡️
PII-gated data flow
14-type detection pipeline scans every outbound message. Per-trust-level policies: BLOCK, REDACT, HASH, or PASS. Adaptive calibration reduces false positives.
📊
Behavioral trust scoring
Formula (0.4×success + 0.2×uptime + 0.2×threat + 0.2×integrity) continuously evaluates peers. Upgrades require history; downgrades are instant.
📋
Compliance built-in
HIPAA, SOC2, GDPR audit trails as compliance modes. Every federation event produces a structured record searchable via HNSW.
🤝
9 MCP tools + 10 CLI commands
Full lifecycle: federation_init, federation_send, federation_trust, federation_audit, and more.
Example: two teams sharing fraud signals without sharing customer data
bash
# Team A: initialize federation and generate keypair
npx claude-flow@latest federation init
# Team A: join Team B's federation endpoint
npx claude-flow@latest federation join wss://team-b.example.com:8443
# Team A: send a task — PII is stripped automatically before it leaves
npx claude-flow@latest federation send --to team-b --type task-request \
--message "Analyze transaction patterns for account anomalies"# Team A: check peer trust levels and session health
npx claude-flow@latest federation status
See issue #1669 for the complete architecture, trust model, and implementation roadmap.
bash
# Claude Code plugin
/plugin install ruflo-federation@ruflo
# Or via CLI
npx claude-flow@latest plugins install @claude-flow/plugin-agent-federation
Claude Code: With vs Without Ruflo
Capability
Claude Code Alone
+ Ruflo
Agent Collaboration
Isolated, no shared context
Swarms with shared memory and consensus
Coordination
Manual orchestration
Queen-led hierarchy (Raft, Byzantine, Gossip)
Memory
Session-only
HNSW vector memory with sub-ms retrieval
Learning
Static behavior
SONA self-learning with pattern matching
Task Routing
You decide
Intelligent routing (89% accuracy)
Background Workers
None
12 auto-triggered workers
LLM Providers
Anthropic only
5 providers with failover
Security
Standard
CVE-hardened with AIDefence
Architecture overview
code
User --> Claude Code / CLI
|
v
Orchestration Layer
(MCP Server, Router, 27 Hooks)
|
v
Swarm Coordination
(Queen, Topology, Consensus)
|
v
100+ Specialized Agents
(coder, tester, reviewer, architect, security...)
|
v
Memory & Learning
(AgentDB, HNSW, SONA, ReasoningBank)
|
v
LLM Providers
(Claude, GPT, Gemini, Cohere, Ollama)
How to grade your agent setup, scan tool configs for security, detect changes between runs, and eject a project into a standalone agent toolkit. The audit-my-setup doc.
v3.8.0 SOTA matrix vs LangGraph / AutoGen / CrewAI on darwin-arm64 + linux-x64. ruflo wins cold start, single turn, RSS by 1.3×–1953×. The is-it-fast doc.
Claude Flow is now Ruflo — named by rUv, who loves Rust, flow states, and building things that feel inevitable. The "Ru" is the rUv. The "flo" is working until 3am. Underneath, powered by Cognitum.One agentic architecture, running a supercharged Rust-based AI engine, embeddings, memory, and plugin system.