Skip to content
Enrollment is open — create your account and start with the free coursesCreate account
MASTERYCERTIFICATIONS
Generative AIIntermediateEnglish

AI Agents with Claude

Build, evaluate and secure production applications and agents on the Claude API, from the first request to tools, MCP and guardrails.

Created by Mastery Certifications Academic Team
30 h29 lessonsUpdated October 2026
Also available in Spanish

What you’ll learn

  • Implement an agent loop with parallel tool calls, error results and safe stop conditions
  • Build an eval harness that compares versions by category and catches regressions
  • Put permission gates, human approval and untrusted-content handling in front of side effects
  • Deliver a documented production-style agent: design, tool schemas, eval plan and safety controls

About this course

Large language models are moving from chat windows into products: assistants that look up real data, take actions and work through multi-step tasks. Building them so that they are reliable, affordable and safe takes more than a good prompt. You need to understand the API, design tools, own or choose an agent loop, and prove with evaluations that the result works.

In this hands-on course you build a support-and-operations agent for a fictional logistics company on the Claude API. You will handle every stop reason and error, write system prompts and structured outputs, use prompt caching, implement the agent loop and the SDK Tool Runner, connect tools through the Model Context Protocol, and compare your own loop with Managed Agents and the Claude Agent SDK. Practices run in your browser against a simulator with the shape of the real SDK; labs on your own computer use the real API.

The course prepares you for Mastery Certified: AI Agent Developer (Claude), an independent certification, with a practice exam and the final project the certification requires. You finish with a documented, production-style agent: its design, tool schemas, eval plan and safety controls.

Course objectives

  • Build Claude applications that handle every stop reason, streaming and API errors correctly
  • Write system prompts, structured outputs and cache-friendly prompts that hold up in production
  • Design tools, implement agent loops and choose between workflows and agents
  • Connect tools and data through the Model Context Protocol and choose an agent platform
  • Evaluate, observe and secure agents against prompt injection and excessive agency

Skills you’ll gain

  • AI Agents
  • LLM Application Development
  • Model Context Protocol (MCP)
  • Prompt Engineering

Course content

6 modules · 29 lessons · 30 h

Try an interactive practiceLab: build the agent loop
01Claude API fundamentals

Requests, responses, models, tokens, stop reasons, streaming and errors: the foundations every Claude application stands on.

02Prompt and context engineering

System prompts that hold up, structured outputs you can parse, long documents and prompt caching.

  • System prompts that hold up in production (Reading)12 min
  • Structured outputs you can parse (Code)24 min
  • Long context, documents and files (Video)14 min
  • Practice: prompt caching and context budgets (Exercise)30 min
  • Knowledge check: prompt and context engineering (Quiz)12 min
03Tool use and agent loops

Design tools, write the agent loop, use the Tool Runner and choose between workflows and agents.

  • Designing tools Claude can use (Code)26 min
  • Lab: build the agent loop (Lab)Preview38 min
  • The Tool Runner and when to own the loop (Reading)14 min
  • Workflows or agents? Choosing the pattern (Video)15 min
  • Practice: routing and evaluator-optimizer (Exercise)28 min
  • Knowledge check: tool use and agent loops (Quiz)13 min
04MCP and integrations

The Model Context Protocol, the MCP connector, and how to choose between your own loop, Managed Agents and the Claude Agent SDK.

  • The Model Context Protocol (Reading)16 min
  • Reading MCP messages (Code)26 min
  • Lab: the MCP connector and client-side MCP (Lab)30 min
  • Agent SDK, Managed Agents or your own loop (Reading)18 min
  • Knowledge check: MCP and integrations (Quiz)12 min
05Evaluation, observability and safety

Evals that catch regressions, traces and costs you can read, and the guardrails an agent needs in production.

  • Building an eval harness (Code)28 min
  • Practice: observability, cost and context (Exercise)34 min
  • Prompt injection and the OWASP Top 10 for LLM applications (Reading)13 min
  • Lab: guardrails and human approval (Lab)32 min
  • Knowledge check: evaluation, observability and safety (Quiz)13 min
06Certification capstone

The exam guide, the final project and a practice exam for the independent certification.

  • Certification exam guide (Reading)10 min
  • Final project: ship Talmar Assist (Project)1 h 30 min
  • Certification practice exam (Exam)45 min

Requirements

  • Working knowledge of Python: functions, dictionaries and lists
  • Basic understanding of HTTP APIs and JSON
  • An Anthropic API key for the labs on your own computer and the final project

Who this course is for

  • Software developers building applications on large language models
  • Data and machine learning engineers moving prototypes into production
  • Technical leads who review and approve AI agent designs

Earn a verifiable certificate

Complete the course requirements to receive a digital certificate with a unique serial number and QR code. Share it on LinkedIn and let anyone confirm it in seconds.

This course prepares you for

Mastery Certified: AI Agent Developer (Claude)

Your instructor

Mastery Certifications Academic Team

Course design, review and quality assurance

The Mastery Certifications academic team designs, reviews and maintains every course in the catalog. Each course is checked by a subject-matter instructor before it is published, and its practice exercises are verified automatically.

Frequently asked questions

Is this an official Anthropic certification?

No. Mastery Certified: AI Agent Developer (Claude) is an independent certification by Mastery Certifications. It is not affiliated with, endorsed by or an official certification of Anthropic. Claude is a product of Anthropic, PBC.

Does the course include the certification exam?

The course prepares you for it with a practice exam and the final project that the certification requires. The graded certification exam is taken separately on the platform.

Do I need an API key?

Not for the practices: they run in your browser against a simulator with the same shape as the Anthropic Python SDK, with no key and no cost. The labs on your own computer and the final project use the real API with your own key.

Related courses

  • enEnglish

    Generative AI

    Generative AI for Business Leaders

    Build an AI strategy that creates measurable value, without betting the business on hype.

    Mastery Certifications Academic Team

    Intermediate14 h20 lessonsAI tutor
  • enEnglish

    Process Automation

    Workflow Automation with n8n

    Design, build and operate production workflows with n8n: APIs, webhooks, error handling, AI agents and self-hosting.

    Mastery Certifications Academic Team

    Intermediate30 h25 lessons
  • enEnglish

    Process Automation

    BPMN Process Automation with Camunda

    Model, orchestrate and operate business processes with BPMN, DMN and Camunda 8, from the first workshop to production.

    Mastery Certifications Academic Team

    Intermediate30 h26 lessons