Build with Claude
For when you want Claude inside your own software. There are four ways in, and picking the wrong one is the most popular mistake in the pond.
Four ways in. Three of them are not the one you should have picked.
Four ways in
| You want to | Use | What you get |
|---|---|---|
| Call Claude directly from your code | Claude API and client SDKs | Direct access to the Messages API. You write the tool loop yourself, or let the SDK's beta tool runner drive it. |
| Embed Claude Code's agent in your own Python or TypeScript app | Agent SDK | A library with Claude Code's built-in tools, permissions, sessions, hooks, subagents and MCP support, running in a process you operate. |
| Have Anthropic host the agent | Managed Agents | A hosted agent harness that runs the loop for you, in Anthropic's cloud sandbox or one on your own infrastructure. |
| Work interactively or script one-off tasks | Claude Code CLI | The terminal tool. claude -p with --output-format json also lets any language drive the same agent loop. |
Clodhopper move: Rebuilding file editing, permissions and a tool loop from scratch on the raw API, then discovering the Agent SDK ships all of it.
The Agent SDK in one screen
The Agent SDK gives you the same tools, agent loop and context management that power Claude Code, in Python and TypeScript.
pip install claude-agent-sdknpm install @anthropic-ai/claude-agent-sdkIt loads skills, commands and memory from your project's .claude/ folder the same way Claude Code does. Authenticate with an API key: Anthropic does not allow third-party products to offer claude.ai login unless previously approved.
The pieces you will hear about
- Tool useClaude asks your code to run a function and uses the result. This is how it does anything beyond writing text.
- MCPThe Model Context Protocol, an open standard for connecting AI tools to outside data and services. Write a server once, use it from Claude Code, the apps and the SDK.
- Agent SkillsPackaged instructions and resources an agent loads when relevant. Anthropic donated the Agent Skills standard to the Agentic AI Foundation on October 22, 2026.
- PluginsBundles of skills, agents, hooks and MCP servers.
- Prompt cachingReuse the stable front of your prompt at a fraction of the input price.
- Batch APIHalf price for work that can wait.
- EffortA setting that steers how much the model thinks. Thinking itself is adaptive on current models.
- Web search and code executionTools that run on Anthropic's side. Web search is billed per search.
Where you can run it
Besides Anthropic's own API, current Claude models are available through Amazon Bedrock, Claude Platform on AWS, Google Cloud and Microsoft Foundry. Model IDs differ slightly on Bedrock, where they carry an anthropic. prefix. See the models page for IDs and the pricing page for rates.
The build shelf of the video library has API, Agent SDK and MCP walkthroughs, and the Bullfrog path puts them in order.
Source: the Agent SDK overview, the models overview and the pricing page.