Claude Agent SDK: Build Code-Aware Agents
The Claude Agent SDK is Anthropic's framework for building production AI agents. Per the official docs, the SDK lets you "build AI agents that autonomously read files, run commands, search the web, edit code, and more." It exposes the same tools, agent loop, and context management that power Claude Code, programmable in Python and TypeScript.
The 30-second version
- What: Anthropic's framework for production AI agents with built-in file, shell, and web tools
- Who: developers building code-aware agents on Claude models
- Price: SDK is free; you pay for Claude API usage (Agent SDK credit starting June 15, 2026 for subscription plans)
- Best at: agents that work on files, run shell commands, search code, and edit it
- Verdict: pick Claude Agent SDK when your agent needs filesystem and shell access out of the box
What the Claude Agent SDK actually is
The Claude Agent SDK is unusual in the category. Most agent frameworks give you the loop and the prompts; you bring your own tools. The Claude Agent SDK ships with a complete tool set out of the box: Read, Write, Edit, Bash, Glob, Grep, WebSearch, WebFetch, and a Monitor tool that watches background scripts and reacts to each output line.
The result is a framework you can use to ship an agent in roughly 10 lines of code. The framework handles tool execution, the agent loop, and context management. You provide the prompt, the SDK does the rest.
Beyond the built-in tools, the SDK supports MCP (Model Context Protocol) servers, so you can plug in browser automation (via Playwright MCP), database access, or any custom MCP integration. Subagents let your main agent spawn specialized agents with focused tools. Sessions let you maintain context across separate invocations.
The TypeScript package bundles a native Claude Code binary as an optional dependency, so you do not need to install Claude Code separately to use the SDK.
Key capabilities
- Python and TypeScript SDKs as first-class
- Built-in tools: Read, Write, Edit, Bash, Monitor, Glob, Grep, WebSearch, WebFetch, AskUserQuestion
- MCP integration for adding any external system
- Subagents for delegating focused subtasks
- Hooks: PreToolUse, PostToolUse, Stop, SessionStart, SessionEnd, UserPromptSubmit
- Sessions for persistent context across invocations
- Permissions system to control which tools an agent can use
- Skills, slash commands, and CLAUDE.md memory support
- Multi-provider authentication: Anthropic API, Amazon Bedrock, Google Vertex AI, Microsoft Azure AI Foundry
Pricing
The SDK is free. You pay for Claude API usage. Starting June 15, 2026, Agent SDK and claude -p usage on subscription plans draws from a new monthly Agent SDK credit, separate from interactive usage limits. The SDK supports authentication via Anthropic API key directly, or via Amazon Bedrock, Google Vertex AI, and Microsoft Azure AI Foundry credentials.
Who the Claude Agent SDK fits
- Developers building agents that work on files (code review, refactoring, content generation)
- Teams shipping production CI/CD agents
- Anyone who wants filesystem + shell + web access without writing the tool wrappers
- TypeScript developers who want a first-class TypeScript agent SDK
Who it does not fit
- Multi-provider stacks (Claude SDK is Anthropic-centric, though Bedrock and Vertex auth options help)
- Voice agents (no Realtime equivalent; OpenAI Agents SDK is better here)
- Pure data ingestion + RAG without file work (LlamaIndex is more direct)
Claude Agent SDK vs OpenAI Agents SDK and LangChain
| Axis | Claude Agent SDK | OpenAI Agents SDK | LangChain |
|---|---|---|---|
| Vendor model | Anthropic only | OpenAI only | Any LLM |
| Languages | Python + TypeScript (both first-class) | Python | Python + JavaScript |
| Built-in tools | Read, Write, Edit, Bash, etc. | You write your own | Large ecosystem of tools |
| MCP support | Native | Via custom integration | Via custom integration |
| Subagents | Native | Handoffs primitive | Via LangGraph |
| Best for | Code-aware long-running agents | OpenAI agents + voice | Multi-provider flexibility |
Full breakdown: Claude Agent SDK vs OpenAI Agents SDK.
Pros and cons
Pros:
- Built-in filesystem and shell tools save hours of glue code
- Python + TypeScript both first-class is rare in the category
- Native MCP integration plugs into hundreds of community MCP servers
Cons:
- Anthropic-only models
- No native voice agent path (OpenAI wins here)
- The filesystem-first design assumes your agent has local file context, which not every use case needs
How I'd use the Claude Agent SDK on SellerShorts
I built our internal content QA agent on the Claude Agent SDK. The agent reads a draft markdown file, runs grep against the banned-word list, suggests edits, and writes the cleaned-up draft. The built-in Read, Edit, and Grep tools meant I wrote roughly 30 lines of TypeScript for the whole thing. The same agent on LangChain would have needed tool wrappers, sandbox setup, and prompt scaffolding. For code-aware agents, Claude Agent SDK has the highest signal-to-noise ratio in 2026.
Try the Claude Agent SDK
Get started with Claude Agent SDK
Free SDK on pip or npm. Pay for Claude API usage as you build.
Building agents for Amazon? See the Amazon AI hub.
Frequently asked questions
What is the Claude Agent SDK?
Per the official Anthropic docs, the Claude Agent SDK lets you 'build AI agents that autonomously read files, run commands, search the web, edit code, and more. The Agent SDK gives you the same tools, agent loop, and context management that power Claude Code, programmable in Python and TypeScript.'
Does Claude Agent SDK support TypeScript?
Yes. Both Python and TypeScript are first-class. The TypeScript package ships a native Claude Code binary as an optional dependency so you don't need to install Claude Code separately.
What built-in tools does the Claude Agent SDK have?
Read, Write, Edit, Bash, Monitor (background script watcher), Glob, Grep, WebSearch, WebFetch, and AskUserQuestion. You can also plug in any MCP server (Playwright, databases, custom APIs) as additional tools.
Claude Agent SDK vs Managed Agents: what is the difference?
Per Anthropic's docs: Agent SDK runs the agent loop inside your own process on your infrastructure. Managed Agents is Anthropic's hosted REST API where Anthropic runs the agent and the sandbox. A common path is to prototype with Agent SDK locally, then move to Managed Agents for production at scale.