agent-sdk-doc

You build autonomous agents by installing the Claude Agent SDK and configuring tool loops within your environment. The documentation provides complete setup guides, API references, and code snippets for both Python and TypeScript to implement production-grade autonomous workflows.

MCP integration connects your subagents to external tools and data sources during their tool loops. The documentation provides tutorials on configuring MCP servers alongside hooks, enabling agents to autonomously fetch context and execute multi-step actions securely.

Yes, the Claude Agent SDK fully supports both Python and TypeScript. The documentation includes official quickstart guides, API references, and migration notes to help you implement autonomous agents in either language within your own environment.

You add persistent context to autonomous agents by utilizing CLAUDE.md-based memory files. The documentation covers how to structure these files to maintain instructions and context across agent runs, ensuring continuous operational state within your workflows.

The best way to track costs and add observability is by using the SDK's built-in monitoring features. The documentation includes tutorials on implementing observability, tracking token usage, and calculating hosting costs for running autonomous agents in production.

Yes, you can use hooks to customize tool loop behavior by injecting custom logic before or after agent actions. The documentation provides specific tutorials on building and registering hooks to intercept, modify, or extend agent workflows dynamically.