Claude Agent SDK
Anthropic's harness for agents that run tools in a loop.
- made by
- Anthropic
Magic Ship is one shop in Vancouver, BC, working remotely with clients worldwide. We are not a partner, reseller, or certified vendor of Anthropic - we just build with this.
What Claude Agent SDK is
The Claude Agent SDK is Anthropic's library for building agents on the same loop that runs Claude Code: the model plans, calls tools, reads the results, and continues until the task is done. It ships file, shell, and search tools, plus subagents, hooks, and permission controls, and speaks MCP for anything else. There are TypeScript and Python bindings.
How we use it
We build on it when the work is file- and command-shaped: repo migrations, test generation, triage across a checkout. Permissions are declared up front so an agent reads widely and writes narrowly, and hooks give us a place to run the project's own lint and test commands before an edit is accepted. Anything client-specific goes in as an MCP server, so the same tools also work from Claude Code and other clients.
Where it is the wrong choice
It ties you to Claude and to a loop Anthropic controls, which is the wrong trade when a client needs to switch model vendors or wants control flow written out as an explicit graph. LangGraph or plain API calls fit that better.
Service lines it turns up in
Related tools
More in Agent frameworks
Other tools in the same service lines
Building something on Claude Agent SDK?
Send the problem rather than a job spec. You get an answer on scope, on fit, and on whetherClaude Agent SDK is even the right call for it.
Start a projectClaude Agent SDK and Anthropic are trademarks of their respective owners, used here to say what we work with.