Ismail-2001/agent-compose
ramework-agnostic multi-agent orchestrator. Declaratively compose LangGraph, CrewAI, and OpenAI SDK agents in a single YAML pipeline for production-ready AI workflows.
Platform-specific configuration:
{
"mcpServers": {
"agent-compose": {
"command": "npx",
"args": [
"-y",
"agent-compose"
]
}
}
}Add the config above to .claude/settings.json under the mcpServers key.
<div align="center">
The Orchestrator for the Agentic Era.
🚀 *One YAML. Any Framework. Infinite Scale.*
[](LICENSE) [](https://python.org) [](https://deepseek.com) []()
---
Build multi-agent AI systems like you build infrastructure with Docker Compose. Stop writing brittle orchestration glue code. Define your topology in YAML and ship.
Quick Start • Why agent-compose? • DeepSeek Mastery • Architecture
</div>
---
Multi-agent development is currently in its "Manual Era." Teams spend 70% of their time writing orchestration logic—handling state, resolving dependencies, and tracking costs.
agent-compose brings order to the chaos. It is the first framework-agnostic orchestrator that allows you to mix LangGraph, CrewAI, and OpenAI SDK agents in a single, declarative pipeline.
Prototype an agent in CrewAI, refine another in LangGraph, and keep a simple generating agent as a raw LLM call. agent-compose handles the data flow, dependency resolution, and parallel execution.
---
pip install agent-compose[all]name: executive-research-pipeline
description: "LangGraph (Research) → CrewAI (Analysis) → DeepSeek (Writing)"
agents:
researcher:
framework: langgraph
model: deepseek-chat
system_prompt: "Thoroughly research the given topic and extract data points."
tools: [weLoading reviews...