Workshops: Agentic Coding Tools in Depth

Claude Agent SDK Workshop: Build Custom Engineering Agents

The Claude Agent SDK gives developers the agent loop, tools and context management that power Claude Code as a library for Python and TypeScript. In two days your team designs, builds and deploys an internal engineering agent, for example for dependency updates, incident triage or code migrations, with the permissions, logging and tests that production use requires.

Who is this workshop for?

  • Backend and platform engineers who want to build agents, not only use them
  • Developer experience teams automating recurring engineering chores
  • AI engineers moving from single model calls to tool-using agents
  • Software architects weighing build versus buy for internal agents
  • Teams whose scripts around Claude Code's headless mode have hit their limits

What you will take away

  • Explain how the agent loop, built-in tools, context management and subagents fit together in the Claude Agent SDK.
  • Add custom tools to an agent, implemented in your own code or connected through MCP servers.
  • Control what the agent may do with permission modes, tool allowlists, approval callbacks and hooks.
  • Manage sessions, resume long-running work and keep context focused on large tasks.
  • Package, deploy, monitor and test an internal agent in an isolated environment with clear cost limits.

Agenda

  1. Module 1

    Day 1: Agent fundamentals

    • From a model call to an agent loop: gather context, act, verify
    • What the SDK provides and what you build yourself
    • Choosing your first use case
  2. Module 2

    Day 1: The first agent in code

    • Project setup in Python or TypeScript, authentication
    • Exercise: a repository analysis agent with read-only tools
    • System prompts, project instructions and settings sources
  3. Module 3

    Day 1: Custom tools and MCP

    • Tools with clear schemas and helpful error messages
    • Exercise: tools for your issue tracker, build system or internal API
    • Reusing existing MCP servers instead of writing everything yourself
  4. Module 4

    Day 2: Permissions and safety

    • Permission modes, allowed and disallowed tools
    • Approval callbacks and hooks for sensitive actions
    • Sandboxing, network limits and untrusted input
  5. Module 5

    Day 2: Sessions, subagents and context

    • Streaming and resuming sessions
    • Subagents for parallel or specialised subtasks
    • Keeping context lean on long tasks
  6. Module 6

    Day 2: Deployment and operations

    • Running agents in containers, CI jobs or as internal services
    • Logging, tracing, cost tracking and evaluation with test tasks
    • Exercise: ship the agent to a staging environment

Which agents are worth building?

Good candidates are recurring engineering tasks with clear inputs and verifiable results: updating dependencies and fixing the resulting test failures, triaging incoming bug reports, migrating code to a new internal API, or answering questions about your platform. We help you pick a case with real value and manageable risk.

Just as important is what not to build: if Claude Code with a good skill or a GitHub Action already solves the problem, a custom agent only adds maintenance.

Agent SDK, Claude Code or a plain API call?

Not every automation needs an agent. A single model call with structured output suits classification and extraction; Claude Code in headless mode suits scriptable tasks inside a repository. The Agent SDK is the right choice when you need your own tools, your own interface or service, fine-grained permission logic and full control over deployment.

Day one produces a working agent; day two makes it safe and ready to operate. Internal systems can be replaced by mock services where needed. neooutput AI Code Engineers support your team at your premises or via video conference, in either English or German, with optional follow-up coaching on the way to production.

Frequently asked questions

Python or TypeScript?

The SDK is available for both. We work in the language your team uses for tooling; mixed teams can pair up in either language.

Can we use the SDK with Amazon Bedrock or Google Vertex AI?

Yes. Like Claude Code, the SDK can authenticate through these cloud providers as well as the Anthropic API. That helps with existing contracts and data residency requirements, and we configure the access you already use.

How is this different from building MCP servers?

An MCP server exposes tools to existing clients such as Claude Code. With the Agent SDK you build the other side: the agent that decides which tools to use, when and how. Many teams need both, and we show how they fit together.

Workshop

Get in touch

Which recurring engineering task should your first in-house agent take over?

Tell us roughly how many people will join and pick dates that work – we will get back to you within 24 hours with a proposal for content, format and schedule.

Prefer to talk first?

15 minutes, free and without obligation – with our AI Trainers and AI Code Engineers.

Book a call for free (opens in new tab) +49 761 519 751 50 (24/7)

Request a training / workshop

Workshop: Claude Agent SDK Workshop: Build Custom Engineering Agents · 2 days

Dates that work for you (optional)

Suggestions about 2 weeks ahead – pick as many as you like

Reply usually within 24 hours · jocapps is ISO 27001 certified · Privacy