This plan assumes you already use Claude regularly and can read code, and that you can give it about an hour on weekdays plus a longer weekend session. Time is allocated roughly in proportion to the blueprint, with extra attention to context management because it trips up so many candidates. If you are new to building with Claude, stretch each week to two.
Before you start
Take the 10-question diagnostic and download the official exam guide from the certification page on the Anthropic Partner Academy. Note your two weakest domains; you will revisit them in week four.
Week 1 — Agentic architecture
- Read our agentic architecture guide and work the sample questions.
- Build a tool-use loop against the Messages API by hand: two tools, one deliberate tool error, an iteration cap and an escalation path.
- Add a coordinator with two parallel subagents that return structured findings with sources.
- Weekend: write down, for each business rule in your build, whether it is enforced by prompt, code, hook or human.
Week 2 — Claude Code and tools
- Set up a repository with a root CLAUDE.md, a subdirectory CLAUDE.md, shared settings with permission rules, one PreToolUse block and one PostToolUse formatter.
- Package a repeated routine as a project command or skill and define one subagent with a narrow tool list.
- Build a small MCP server with one tool and one resource over stdio; rewrite its tool descriptions until Claude always picks the right tool.
- Read our Claude Code and tool design guides.
Week 3 — Prompts, structured output and context
- Build an extractor with a forced tool, a nullable field and an enum; add a validation-retry loop that feeds back the exact error.
- Create a ten-item evaluation set and measure two prompt versions.
- Lay out a long prompt for caching, then break the cache by changing the prefix and observe the difference in usage numbers.
- Write backoff-with-jitter retry logic and a per-item status table for a batch job.
- Read the prompt and context management guides.
Week 4 — Practice under exam conditions
- Day 1: a full, timed 60-question practice exam in one sitting. Aim for two minutes per question and flag rather than stall.
- Days 2–4: review every missed question, then reread the guide for your two weakest domains and redo their builds.
- Day 5: a second practice attempt. Questions and options are reshuffled, but many questions repeat, so judge yourself on whether you can explain each answer, not on the score alone.
- Final days: skim the exam guide's task statements and check you can explain each in a sentence.
Free resources worth your time
Anthropic Academy offers free self-paced courses, including Claude 101, Claude Code 101, Building with the Claude API, Claude Code in Action, Introduction to Model Context Protocol, Model Context Protocol: Advanced Topics and Introduction to Agent Skills. The Claude documentation covers tool use, prompt caching, batch processing and Claude Code configuration in depth. Our resources page links to each.
Signals that you are ready
You consistently score well above 70% on fresh practice forms, no domain sits far below the others, you finish with time to spare, and — most importantly — you can explain why each wrong option is wrong. Practice scores are a readiness signal, not a prediction of your scaled result.