# @sparkleideas/agentic-flow [Health: Active]

**Category:** 🤖 Coding Agents  
**Repository:** https://github.com/ruvnet/agentic-flow  
**GitHub Stars:** 830  
**npm Downloads (last month):** 415  
**Views:** 0  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/sparkleideas-agentic-flow

## Description
Production-ready AI agent orchestration platform with 66 specialized agents, 213 MCP tools, ReasoningBank learning memory, quantum-resistant Jujutsu VCS, and autonomous multi-agent swarms. Built by @ruvnet with Claude Agent SDK, neural networks, memory pe

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `npx` (confidence: high):

```json
"mcpServers": {
  "agentic-flow": {
    "command": "npx",
    "args": ["-y","agentic-flow"]
  }
}
```

## Documentation

## What @sparkleideas/agentic-flow does

@sparkleideas/agentic-flow organizes coding work across specialized agents and exposes the orchestration layer through MCP. Its feature set includes task routing, self-learning hooks, background workers, adaptive memory, and swarm coordination. The project describes 66 specialized agents and 213 MCP tools, covering roles such as coding, testing, reviewing, planning, research, coordination, and performance analysis.

The platform also includes ReasoningBank-based learning, GNN-enhanced context, attention mechanisms, and SONA for adaptive learning. These capabilities are intended to help the system learn from prior task outcomes, select agents, and coordinate multiple agents during execution.

## How it works

A typical workflow begins by initializing the project:

```bash
npx agentic-flow init
npx agentic-flow hooks pretrain
claude
```

Initialization prepares the workspace, while `hooks pretrain` bootstraps intelligence from the codebase. Once the hooks are active, tasks can be routed to an appropriate agent with a command such as:

```bash
npx agentic-flow hooks route "implement user authentication"
```

The routing result identifies an agent and reports a confidence value. Learning metrics are available through `hooks metrics`. Keyword-triggered background workers can be dispatched with commands such as `workers dispatch "ultralearn how caching works"`.

For MCP use with Claude Code, start the server with:

```bash
npx agentic-flow mcp start
```

Applications can also import `AgenticFlow` from the package, initialize it, and call `route()` from TypeScript.

## Setup and configuration

The documented quick-start path requires Node.js 18 or newer and the `agentic-flow` package available through `npx`. Run the initialization command from the project you want to analyze, then pretrain the hooks against that codebase. The server can be started separately with the MCP command.

These profiles represent different trade-offs among latency, throughput, learning quality, and memory usage. The Balanced profile is described as the default.

## Tools and capabilities

- Route tasks to specialized agents and inspect routing confidence.
- Provide self-learning hooks that record and reuse task patterns.
- Dispatch background workers for operations such as ultralearning, optimization, and auditing.
- Coordinate autonomous multi-agent swarms through MCP tools.
- Use ReasoningBank memory and cross-agent pattern sharing.
- Select among attention and routing strategies, including flash, multi-head, linear, hyperbolic, and mixture-of-experts approaches.
- Expose 213 MCP tools for orchestration and swarm coordination.
- Integrate programmatically through the `AgenticFlow` TypeScript API.

The listed agent roles include coder, reviewer, tester, planner, researcher, hierarchical coordinator, mesh coordinator, adaptive coordinator, issue tracker, release manager, and performance analyzer.

## Limitations and notes

The quick-start workflow is centered on Claude Code. SONA and AgentDB capabilities are presented as part of the v2 platform, with some components described using alpha release labels. Performance figures in the project materials are benchmark claims tied to particular runtimes, profiles, or workloads rather than general guarantees.

_Full upstream README: https://allmcps.com/mcp/sparkleideas-agentic-flow/readme_