anthropic-agent-sdk

anthropic-agent-sdk

Claude Agent SDK (code.claude.com/docs/en/agent-sdk) — the official SDK for building agents in TypeScript / Python reference.query() agent loop execution by, custom tools (in-process MCP), external MCP connections, hooks, subagents, permissions (canUseTool), session persistence/resume/fork, streaming, structured outputs, covers production hosting.

For the Claude Code CLI itself, see anthropic-claude-code / anthropic-claude-code-extend; for calling the Messages API directly, see anthropic-api-core; for tool use / Agent Skills / MCP on the Claude API side, see anthropic-api-tools-mcp (this skill handles the side that builds and operates agents via the SDK).

Directory structure

skills/anthropic-agent-sdk/
  SKILL.md
  references/
    getting-started/
      README.md
      agent-loop.md
      claude-code-features.md
      migration-guide.md
      overview.md
      quickstart.md
    api-reference/
      README.md
      python.md
      typescript.md
    extending/
      README.md
      custom-tools.md
      hooks.md
      mcp.md
      plugins.md
      skills.md
      slash-commands.md
      subagents.md
      tool-search.md
    sessions-io/
      README.md
      file-checkpointing.md
      session-storage.md
      sessions.md
      streaming-output.md
      streaming-vs-single-mode.md
      structured-outputs.md
      todo-tracking.md
    control/
      README.md
      modifying-system-prompts.md
      permissions.md
      user-input.md
    operations/
      README.md
      cost-tracking.md
      hosting.md
      observability.md
      secure-deployment.md
      troubleshooting.md
  samples/
    README.md
    minimal-agent.md
    custom-tools.md
    mcp-servers.md
    hooks.md
    permissions.md
    subagents.md
    sessions.md
    streaming.md
    structured-outputs.md
    system-prompts.md
  scripts/
    README.md
    setup.md
    runtime.md

Exploration procedure

Derive the category from the task, and identify the target page in the category's README.md:

1. Find the category corresponding to the task in the mapping table below
2. of that category references/{category}/README.md（samples/ scripts/ is the README.md directly under it) to identify the target page
3. Read the .md of the relevant page to check the details

Task → Category mapping

Task	Category	Reference README
Want to know an overview of the Agent SDK and how the agent loop works / want to install it and build a first agent / migration from the Claude Code SDK / want to use Claude Code features (CLAUDE.md, Agent Skills, hooks) within the SDK	getting-started	references/getting-started/README.md
query() や ClaudeAgentOptions / Options such as the complete types and function signatures for TypeScript / Python	api-reference	references/api-reference/README.md
custom tools（in-process MCP, createSdkMcpServer) / want to connect to external MCP servers / want to control tool calls with hooks / want to handle plugins / SKILL.md / slash commands / subagents / tool search	extending	references/extending/README.md
Want to handle session persistence/resume/fork (resume / fork) / saving to external storage / file checkpointing / todo management / streaming output / structured outputs	sessions-io	references/sessions-io/README.md
Want to customize the system prompt, permission mode, or canUseTool want to control permissions with / want to handle user input/approval flows	control	references/control/README.md
Want to know about hosting in production environments / secure deployment / observability with OpenTelemetry / cost/token tracking / troubleshooting errors	operations	references/operations/README.md
Want to know typical usage (minimal agent, custom tools, MCP server connection, hooks, permissions, subagents, session continuation, streaming, structured outputs, system prompt customization)	samples	samples/README.md
Want to know SDK installation / API key configuration / telemetry environment variables for hosting	scripts	scripts/README.md