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.

Ruflo ships a full MCP 2025-11-25 compliant server that exposes 313 tools across five groups: coordination, memory, monitoring, GitHub, and system. It’s the primary integration point for Claude Code, Cursor, Windsurf, and any other MCP-compatible client. The server starts in under 400ms and supports stdio (default), HTTP, and WebSocket transports — pick the one that fits your editor.

Starting the Server

# Standard start — stdio transport, used by Claude Code and Codex
npx ruflo@latest mcp start

Transports

TransportUse CaseCommand
stdioClaude Code, Codex (default)mcp start
httpWindsurf, Cursor, REST clientsmcp start --transport http
websocketReal-time streaming appsmcp start --transport websocket
The stdio transport is the canonical form for Claude Code. The server reads JSON-RPC messages from stdin and writes responses to stdout — no network port is opened. Use http or websocket for editors that connect over a network socket or for remote deployments.

Registering with Editors

Add Ruflo as a named MCP server with a single command:
claude mcp add ruflo -- npx ruflo@latest mcp start
Verify the registration:
claude mcp list
Once added, all 313 Ruflo MCP tools are available directly in your Claude Code sessions — swarm_init, agent_spawn, memory_search, hooks_route, and the rest.

MCP Tool Groups

The 313 tools are organized into five groups. Load only what your workflow needs.
GroupKey ToolsApprox. Count
coordinationswarm_init, agent_spawn, task_orchestrate~50
monitoringswarm_status, agent_list, agent_metrics~40
memorymemory_store, memory_search, memory_usage, neural_train~60
githubgithub_swarm, repo_analyze, pr_enhance, code_review~80
systembenchmark and utility tools~83
Set groups at startup:
# Development — coordination + memory only
CLAUDE_FLOW_TOOL_GROUPS=coordination,memory npx ruflo@latest mcp start

# All groups
npx ruflo@latest mcp start
Or set them permanently in claude-flow.config.json:
{
  "mcp": {
    "port": 3000,
    "toolGroups": ["coordination", "memory", "monitoring"]
  }
}

HTTP Transport Configuration

For Windsurf, Cursor (HTTP mode), REST clients, or any scenario where you need a network-accessible MCP endpoint, configure the server programmatically using @claude-flow/mcp:
import { createMCPServer } from '@claude-flow/mcp';

const server = createMCPServer({
  transport: 'http',
  host: 'localhost',
  port: 3000,
  corsEnabled: true,
  corsOrigins: ['http://localhost:8080'],
  auth: {
    enabled: true,
    method: 'token',
    tokens: ['secret-token'],
  },
}, logger);

await server.start();
// HTTP endpoint: http://localhost:3000
// WebSocket endpoint: ws://localhost:3000/ws
The same options are available via the CLI:
npx ruflo@latest mcp start \
  --transport http \
  --port 3000 \
  --cors-origins http://localhost:8080
Set CLAUDE_FLOW_AUTH_TOKEN in your environment to enable token authentication without modifying code:
CLAUDE_FLOW_AUTH_TOKEN=your-secret-token npx ruflo@latest mcp start --transport http

Server Management Commands

1

Check server status

npx ruflo@latest mcp status
2

Stop the server

npx ruflo@latest mcp stop
3

List tools by category

# List all coordination tools
npx ruflo@latest mcp tools --category coordination

# List all tools across all categories
npx ruflo@latest mcp tools
4

View server logs

npx ruflo@latest mcp logs
5

Restart the server

npx ruflo@latest mcp restart

Performance Targets

The @claude-flow/mcp package is built to these targets:
MetricTarget
Server startup<400ms
Tool registration<10ms per tool
Tool execution overhead<50ms
Connection acquire<5ms
Startup time: The full Ruflo CLI cold-start is ~35s on a fresh npx pull due to the ML/embedding dependencies. For plugin scripts that invoke memory commands in hooks, set CLI_CORE=1 to use @claude-flow/cli-core instead — this drops cold-cache wall time from ~35s to ~1.5s (22.9× speedup). The full MCP server always uses the complete CLI regardless of CLI_CORE.After the first invocation, npx caches the package and subsequent starts take ~3s.

MCP Methods Supported

Ruflo’s MCP server is fully MCP 2025-11-25 compliant and implements the complete protocol surface:
MethodDescription
initializeInitialize connection and negotiate capabilities
tools/listList available tools (with optional pagination)
tools/callExecute a tool by name
resources/listList resources with cursor-based pagination
resources/readRead resource content
resources/subscribeSubscribe to resource updates
prompts/listList prompt templates
prompts/getGet a prompt with arguments
tasks/statusGet async task status
tasks/cancelCancel a running task
logging/setLevelSet server log level
pingKeep-alive

Next Steps

Full MCP Tool Reference

Browse all 313 MCP tools organized by group, with parameter schemas and usage examples.

Environment Variables

Configure CLAUDE_FLOW_TOOL_GROUPS, CLAUDE_FLOW_MCP_PORT, CLAUDE_FLOW_AUTH_TOKEN, and more.

Build docs developers (and LLMs) love