OpenAI Agents SDK (Python)
The OpenAI Agents SDK enables you to build agentic AI apps in a lightweight, easy-to-use package with very few abstractions. It's a production-ready upgrade of our previous experimentation for agents, Swarm. The Agents SDK has a very small set of primitives:
- Agents, which are LLMs equipped with instructions and tools
- Handoffs, which allow agents to delegate to other agents for specific tasks
- Guardrails, which enable the inputs to agents to be validated
The Agents SDK has support for MCP.
This enables you to use a wide range of MCP servers to provide tools to your Agents.
Agents - OpenAI API
Agents represent systems that intelligently accomplish tasks,
ranging from executing simple workflows to pursuing complex, open-ended objectives.
New tools for building agents | OpenAI
For Server Developers - Model Context Protocol
modelcontextprotocol/servers: Model Context Protocol Servers
As explained by ChatGPT:
OpenAI Agents SDK vs LangChain
- OpenAI Agents SDK is like a "first-party" minimalistic, efficient way to build powerful agents for OpenAI models.
- LangChain is like a "general-purpose Swiss Army knife": works with OpenAI and others, but has more layers and can get heavy for simple apps.
- CrewAI, Autogen, LlamaIndex — more specialized: teams of agents, multi-turn logic, document-based reasoning.
Use OpenAI Agents SDK if you're mainly using OpenAI models and want clean agents with memory + tools fast.
Use LangChain if you need cross-model, cross-data-source orchestration and don’t mind complexity.
Use CrewAI/Autogen if you're building multi-agent collaboration (e.g., AI project managers + AI workers).
