Claude Agent SDK
Anthropic · TypeScript, Python
The harness behind Claude Code as a library: tool use, MCP servers, Agent Skills and subagents
libraries
What agents are built with. An agent can’t call a library — it runs inside one — so these aren’t listings. The directory lists the services they help an agent call, and /protocols explains the standards they speak.
Run the loop: call a model, pick a tool, call it, repeat.
Anthropic · TypeScript, Python
The harness behind Claude Code as a library: tool use, MCP servers, Agent Skills and subagents
OpenAI · Python, TypeScript
Agents, handoffs between them, guardrails and tracing, with MCP servers as tools
Google · Python, Java
Open-source framework for building and deploying agents, with A2A for talking to other agents
LangChain · Python, TypeScript
Agents as graphs of steps with durable state, so long runs survive restarts and can pause for a person
Mastra · TypeScript
TypeScript agent framework with workflows, memory, evals and a deployable server
Vercel · TypeScript
One TypeScript interface over many model providers, with tool calling, agents and MCP clients
Keep what an agent learned between runs.
LangChain · Python
Extracts, updates and retrieves semantic, episodic and procedural memories for an agent
Stop and ask a person before an agent does something it shouldn't do alone.
HumanLayer · Python, TypeScript
Approvals, escalations and review steps an agent blocks on until a person answers
Let an agent act for a user against other people's APIs without holding their password.
Okta · TypeScript, Python
Token vault, asynchronous user approval and Cross App Access for agents calling third-party APIs