# open-multi-agent — Task DAG Orchestration with MCP

> open-multi-agent turns a goal into a task DAG and orchestrates multiple agents in TypeScript, with MCP support and runnable example scripts.

## Install

Save as a script file and run:

## Quick Use

1. Install / run:
   ```bash
   npm install @open-multi-agent/core
   ```
2. Start / smoke test:
   ```bash
   npx tsx examples/basics/team-collaboration.ts
   ```
3. Verify:
   - Run one example script and confirm it builds a task DAG and completes the workflow with traceable steps.

## Intro

open-multi-agent turns a goal into a task DAG and orchestrates multiple agents in TypeScript, with MCP support and runnable example scripts.

- **Best for:** TypeScript teams who want a lightweight, dependency-minimal multi-agent orchestrator that can trace runs and speak MCP
- **Works with:** Node.js tooling + TS runtime; MCP integrations per repo README
- **Setup time:** 12 minutes

## How to Use It Well

- Setup time ~12 minutes (npm install + run one `tsx` example)
- Goal → task DAG is the core abstraction; you can inspect the DAG to debug agent drift
- GitHub stars + forks (verified): see Source & Thanks

Multi-agent systems fail when state gets fuzzy. A task DAG helps: you can see what’s planned, what ran, what failed, and what’s left. Keep tools read-only first, add write actions with explicit approvals.

### FAQ

**Q: Do I need MCP to use it?**
A: Not strictly. MCP is an integration surface; you can start locally and add MCP tools when needed.

**Q: How do I debug failures?**
A: Treat the task DAG as your source of truth: inspect nodes, inputs/outputs, and rerun only the failing steps.

**Q: Is it heavyweight?**
A: The README emphasizes minimal runtime dependencies; validate your own constraints and keep agents scoped.

## Source & Thanks

> Source: https://github.com/open-multi-agent/open-multi-agent
> License: MIT
> GitHub stars: 6,095 · forks: 2,349

---

<!-- ZH -->

## 快速使用

1. 安装 / 运行：
   ```bash
   npm install @open-multi-agent/core
   ```
2. 启动 / 冒烟测试：
   ```bash
   npx tsx examples/basics/team-collaboration.ts
   ```
3. 验证：
   - Run one example script and confirm it builds a task DAG and completes the workflow with traceable steps.

## 简介

open-multi-agent 在 TypeScript 中把目标自动分解为任务 DAG，并编排多个 agent 协作执行；支持 MCP 与 live tracing。提供 npm 包与示例脚本，适合工程化地跑“多步任务”。

- **适合谁：** 希望在 TypeScript 里用轻依赖的方式编排多 agent，并可追踪运行且兼容 MCP 的团队
- **可搭配：** Node.js 工具链 + TS 运行环境；MCP 集成以 README 为准
- **准备时间：** 12 分钟

## 实战建议

- 上手约 12 分钟（npm 安装 + 跑一个 `tsx` 示例）
- 核心抽象是“目标 → 任务 DAG”，可通过 DAG 结构排查 agent 漂移与遗漏
- GitHub stars + forks（已核验）：见「来源与感谢」

多 agent 容易失败在“状态不清”。任务 DAG 能显式化：计划了什么、跑了什么、失败在哪里、还剩什么。建议先只读工具跑通，再把写操作加上明确审批门。

### FAQ

**必须用 MCP 吗？**
A: 不强制。MCP 是一种集成接口；你可以先本地跑通，再按需接入 MCP 工具。

**失败了怎么排查？**
A: 把任务 DAG 当作真相来源：检查每个节点的输入/输出，只重跑失败步骤。

**会不会很重？**
A: README 强调运行时依赖很少；但仍建议你控制 agent 范围并按需引入工具。

## 来源与感谢

> Source: https://github.com/open-multi-agent/open-multi-agent
> License: MIT
> GitHub stars: 6,095 · forks: 2,349


---
Source: https://tokrepo.com/en/workflows/open-multi-agent-task-dag-orchestration-with-mcp
Author: Agent Toolkit