The OpenAI Agents SDK is a small library for building agents: an agent loop, tool calling, structured handoffs between agents, guardrails, and tracing. OpenAI ships it as the production-ready successor to its experimental Swarm project. The Python package and a separate JavaScript/TypeScript port are both MIT-licensed. Among the vendor SDKs it sits at the low-abstraction end: a few primitives rather than an opinionated framework. It is provider-agnostic in the sense that the README documents support for the OpenAI Responses and Chat Completions APIs "as well as 100+ other LLMs" reached through LiteLLM, though OpenAI models are the default and best-tested path. Multi-agent work is expressed as handoffs (one agent delegating to another) and agents-as-tools. For open-source AI the relevant facts are that the code is MIT and MCP support is native (the SDK understands stdio, Streamable HTTP, and SSE transports), so an agent built on it talks to the same tool servers as any other MCP client. It is still pre-1.0 in version numbering (0.17.x as of May 2026) but is OpenAI's recommended agent-building path and is widely deployed.
The Stack · Agents · Open source
OpenAI Agents SDK
OpenAI's lightweight agent SDK and the production successor to Swarm; MIT, Python and TypeScript, provider-agnostic via LiteLLM, handoffs, native MCP.
Sources
- OpenAI Agents SDK documentation https://openai.github.io/openai-agents-python/
- openai-agents-python LICENSE (MIT) https://raw.githubusercontent.com/openai/openai-agents-python/main/LICENSE
- openai-agents-python README (provider-agnostic, handoffs) https://raw.githubusercontent.com/openai/openai-agents-python/main/README.md
- OpenAI Agents SDK MCP docs https://openai.github.io/openai-agents-python/mcp/
- openai-agents-python releases https://github.com/openai/openai-agents-python/releases
Want a follow-up? Ask the chat about OpenAI Agents SDK in context. It will compare to siblings at the same layer and ground every claim in the wiki.
Other projects at the Agents layer
19 siblings · ordered open first
- Goose (Block) Open source
Block's open agent; under the Linux Foundation's AAIF since Dec 2025; MCP-native; BYOK; extension-based.
- OpenHands Open source
Formerly OpenDevin; sandboxed open coding agent; strong SWE-bench Verified performance.
- Aider Open source
Paul Gauthier's terminal-native open coding agent; BYOK; git-aware diffs; widely loved for its directness.
- Cline Open source
VS Code-native open coding agent; large community of users; began as Claude Dev.
- Continue Open source
Open IDE coding agent for VS Code and JetBrains; broader IDE integration than most.
- LangChain Open source
Most-used open agent framework; LLM tools, chains, RAG primitives, and LangGraph for low-level agent orchestration. MCP via the langchain-mcp-adapters package.
- LlamaIndex Open source
Open framework for data-aware LLM applications; strong on RAG, now also an agent framework via AgentWorkflow. MCP via the llama-index-tools-mcp package.
- AutoGen Open source
Microsoft's multi-agent conversation framework, now maintenance-mode; consolidated with Semantic Kernel into the Microsoft Agent Framework (1.0 GA April 2026).
- Claude Agent SDK Open source
Anthropic's agent SDK extracted from Claude Code; the package is MIT but its use is governed by Anthropic's Commercial Terms, and it is Claude-only. Native plus in-process MCP, subagents. Renamed from the Claude Code SDK September 2025.
- Google ADK (Agent Development Kit) Open source
Google's open agent framework; Apache-2.0, multi-language (Python, Java, Go, TypeScript, Kotlin), Gemini-optimized but model-agnostic, native bidirectional MCP. 2.0 GA May 2026.
- Vercel AI SDK Open source
Vercel's TypeScript AI toolkit; Apache-2.0, model-agnostic via a unified provider API, subagents, native MCP. AI SDK 6 GA December 2025.
- Strands Agents Open source
AWS's open agent SDK; Apache-2.0, Python (TypeScript in preview), Bedrock-default but model-agnostic, native MCP, workflow/graph/swarm multi-agent. 1.0 GA July 2025.
- Microsoft Agent Framework Open source
Microsoft's open agent framework consolidating AutoGen and Semantic Kernel; MIT, Python and .NET, model-agnostic via connectors, native MCP and A2A, graph-based orchestration. 1.0 GA April 2026.
- Pydantic AI Open source
Pydantic's MIT agent framework; Python, model-agnostic, type-safe through Pydantic models, native MCP, A2A plus graph multi-agent. 1.0 released September 2025.
- CrewAI Open source
Independent MIT agent framework; Python, model-agnostic via LiteLLM, role-based "crews" for multi-agent work, MCP via the crewai-tools adapter. 1.0 GA October 2025.
- Mastra Open source
Independent TypeScript agent framework from the Gatsby team; Apache-2.0 core (separate enterprise license for ee/ directories), built on the Vercel AI SDK, native MCP, supervisor/sub-agents. 1.0 released January 2026.
- Claude Code (Anthropic) Proprietary
Anthropic's first-party CLI coding agent; closed-source; subscription model; the dominant closed coding agent in 2026.
- Cursor Proprietary
Closed VS-Code-fork IDE with integrated agentic features; strong developer adoption.
- Devin (Cognition Labs) Proprietary
Cognition's closed autonomous-software-engineer agent; was first to publicize SWE-bench scores.