Complete onboarding reference for every role. Covers Claude Chat, Projects, Artifacts, MCP, Memory, Skills (with 5 workflow patterns + troubleshooting), Claude Code (35 power tricks), Coworker, the 10-step mastery plan, and the full Anthropic Academy catalog with all 17 courses.
The most important first step: understand which surface to use for which task. Most people underutilize Claude because they use one surface for everything.
Browser-based conversational workspace. Brainstorm, analyze documents, build interactive artifacts, run deep research, connect MCP apps. Your primary daily driver.
Agentic coding CLI in your terminal. Full codebase context, file editing, bash execution, git automation, MCP servers, subagents, hooks, and custom skills.
Desktop-level agentic AI for knowledge workers. Multi-step file management, report generation, app integrations — via visual approval flow. Supports scheduled overnight tasks.
Claude Coworker only runs inside the desktop app (Mac or Windows — not the web version). Pro plan required: $20/mo or $17/mo billed annually.
Complex reasoning, system design, security reviews, architecture decisions. Use /effort max. API: claude-opus-4-6
Best speed/quality balance. Implementation, refactoring, feature work. Default choice for most tasks. API: claude-sonnet-4-6
Fastest, most cost-efficient. Quick bug fixes, small repetitive tasks, high-volume pipelines. API: claude-haiku-4-5-20251001
Available at claude.ai on web, iOS, Android, and desktop. No installation. No coding. Your primary interface for everything that doesn't require terminal or file system access.
Upload PDFs, Word docs, spreadsheets, images, code files. Supports up to 1 million token context — roughly 75,000 lines of code or dozens of PDFs in one pass.
Real-time search and article fetching built in. Toggle from the toolbar. Claude searches automatically when queries need current info. Available on Free plan.
Multi-step agent: 10–20 searches, reads full articles, synthesizes a comprehensive report. Competitive intelligence, literature reviews, market analysis. Pro / Max
Python, bash, Node.js in a sandboxed Ubuntu container. Installs packages, creates downloadable files, generates charts. Enable via "Code Execution and File Creation" in the toolbar.
Projects eliminate the cold-start problem — re-explaining context at the beginning of every conversation. Each project has its own system prompt, document library, and searchable conversation history across all sessions.
Standing system prompt for the project — role, output format, domain rules. Applied to every conversation automatically. The single highest-ROI setup step.
Upload PDFs, API docs, style guides, CSVs, architecture diagrams. Claude references them in every conversation. Ideal for recurring reference material you would otherwise re-paste daily.
All chats are searchable. Ask "what did we decide about the ClickHouse schema last week?" and Claude finds it. History is scoped to the project so context stays focused.
Share projects with teammates on Team/Enterprise — shared instructions, documents, and history. Turns Claude from a personal tool into a team multiplier. Team / Enterprise
Create a parent folder "Claude-Cowork" with 4 subfolders: About Me (identity + writing rules), Projects (one subfolder per live project), Templates (your best past work as reusable structure), Claude Outputs (the ONLY folder Claude ever writes to — everything else stays read-only).
A live side panel that renders generated content next to your conversation. Available on all plans. Shareable via public link — no Claude account needed to view.
Full interactive applications in a sandboxed iframe. Recharts, D3, Three.js, Tailwind. Can call the Anthropic API directly ("Claude in Claude") for AI-powered artifacts with MCP integrations.
Architecture diagrams, flowcharts, sequence diagrams, ER diagrams rendered natively. Essential for system design documentation and technical communication.
Reports, READMEs, articles, proposals rendered cleanly. Download or share as a public URL your team can open in a browser — no login required.
Artifacts save state across sessions via a key-value API — up to 5 MB per key. Build trackers, journals, leaderboards, and collaborative tools that remember data between visits.
"Claude in Claude" — artifacts call the Anthropic API at runtime, building fully functional AI apps with MCP integrations running inside your chat panel.
Publish any artifact as a shareable link. Recipients see a live rendered version. Great for prototypes, reports, demos, and sharing interactive tools.
Open standard that lets Claude connect to external services — databases, APIs, SaaS apps — so it can read and act on real data. One click to connect in Claude.ai, one command in Claude Code.
MCP provides the professional kitchen — access to tools, ingredients, and equipment (your real data and services). Skills provide the recipes — step-by-step instructions for how to create something valuable with that kitchen. Together, they enable complex tasks without users needing to figure out every step themselves.
| Connector | What Claude can do |
|---|---|
| Gmail | Search, read, draft, send emails |
| Google Calendar | Read/create events, summarize schedules |
| Google Drive | Search, fetch, and read Drive files |
| Notion | Read/write pages, databases, comments |
| GitHub | List PRs, read issues, manage repos |
| Jira / Asana / Linear | Create, update, query project tickets |
| Slack | Read channels, post messages, search threads |
| Figma / Salesforce | Design file access, CRM data operations |
Connectors = Claude acts inside your apps. Plugins = you do the work while Claude assists from a sidebar. Both in Settings → Integrations. Start with one — master it first, then expand.
# Add GitHub MCP server
claude mcp add github -- npx @modelcontextprotocol/server-github
# List connected servers claude mcp list
# Manage mid-session (no restart needed)
/mcp
# MCP tools as namespaced slash commands
/mcp__github__list_prs
/mcp__slack__send_messageTwo memory layers — automatic summaries from past conversations, and explicit facts you ask Claude to store. Both persist across sessions. Viewable at Settings → Capabilities → Memory.
Claude summarizes things you have shared — role, projects, tools, preferences. Updates automatically. View and delete individual items in Settings at any time.
Trigger with: "remember that I now use ClickHouse" or "forget about my Form Energy prep." Written to persistent memory immediately — up to 30 edits, 100k characters each.
Ask "what did we decide about the Go gateway?" Claude searches full conversation history — scoped to the current Project or outside any Project depending on context.
Memory fully disabled. Nothing stored, summarized, or surfaced. Use for completely stateless, private sessions with no persistence of any kind.
about-me.md — Who you are, what you do, current priorities, what matters now. One well-crafted file beats 50 random uploads. Be intentional — don't dump everything.
anti-patterns.md — Every phrase Claude must NEVER use: "boundaries", "delve", "it's worth noting", "certainly!", "I'd be happy to". This single file dramatically improves output quality across all tasks.
Reusable packaged workflows — teach Claude once how to handle a task, and it applies that knowledge consistently across every future conversation. An open portable standard across Claude.ai, Claude Code, and the API.
Instead of re-explaining your preferences, processes, and domain expertise in every conversation, skills let you teach Claude once and benefit every time. Skills are most powerful for repeatable workflows: generating frontend designs from specs, conducting research with consistent methodology, creating documents that follow your team's style guide, orchestrating multi-step processes that coordinate MCP servers.
Folder: kebab-case only. SKILL.md is case-sensitive. No README.md inside the skill folder.
Level 1 — YAML frontmatter: Always loaded. Tells Claude when to use this skill. Minimizes token usage.
Level 2 — SKILL.md body: Loaded when skill is relevant. Full instructions and guidance.
Level 3 — Linked files: references/ and scripts/ loaded only as needed. Keep SKILL.md under 5,000 words.
.claude/skills/name/SKILL.md (project) or ~/.claude/skills/ (global)/v1/skills endpoint and container.skills param. Admins: deploy workspace-wide via Admin console.---
name: your-skill-name
description: Analyzes Figma design files and generates
developer handoff documentation. Use when user uploads
.fig files, asks for "design specs", "component
documentation", or "design-to-code handoff".
license: MIT
allowed-tools: "Bash(python:*) WebFetch"
metadata:
author: Your Name
version: 1.0.0
mcp-server: your-service
---name: kebab-case, no spaces, no "claude"/"anthropic" prefix (reserved). description: max 1024 chars. No XML angle brackets anywhere (security restriction).
"Manages Linear project workflows including sprint planning, task creation, and status tracking. Use when user mentions 'sprint', 'Linear tasks', 'project planning', or asks to 'create tickets'."
"Helps with projects." → too vague, no trigger phrases
"Does frontend things." → no file types, no user phrasing
Debug: ask Claude "when would you use the [skill name] skill?" — it quotes the description back.
Generate consistent output: documents, presentations, apps, designs. No external tools required — uses Claude's built-in capabilities. Examples: docx, pptx, xlsx, frontend-design skills.
Multi-step processes with consistent methodology. May coordinate across multiple MCP servers. Step-by-step workflow with validation gates. Examples: skill-creator, code-review-agents.
Add workflow knowledge on top of MCP tool access — turn raw API calls into guided, reliable workflows. Embed domain expertise, best practices. Examples: sentry-code-review, context7.
Use when users need multi-step processes in a specific order with dependencies between steps.
Use when workflows span multiple services — design-to-dev handoffs, cross-system reporting.
Use when output quality improves through iteration — reports, content, complex code generation.
Use when same outcome requires different tools depending on context — file type, size, or role.
Use when the skill adds specialized knowledge beyond tool access — compliance rules, financial logic, safety checks. Domain expertise embedded in logic, not tool access.
MCP skill pulling version-specific docs from source repos into Claude's context. Instead of outdated training data, Claude gets live, accurate API docs for any library you are working with.
Official Anthropic skill. Stop-hook pattern enabling multi-hour autonomous coding sessions. Automatically resets Claude's context between tasks. Prevents infinite loops and API overuse.
Adds long-term memory to Claude Code. Carry context and preferences across coding sessions. Persistent memory compression system. #1 GitHub trending repo on launch day.
Most popular Claude skill with 96K+ installs. Transforms Claude's UI output into distinctive, production-grade interfaces with bold typography, unique color palettes, and creative layouts.
Invalid file name: Must be exactly SKILL.md — case-sensitive. skill.md and SKILL.MD both fail.
Bad YAML: Missing --- delimiters, unclosed quotes, XML angle brackets.
Bad folder name: Must be kebab-case. No spaces, no capitals, no underscores.
Description is too generic. Add specific trigger phrases users would actually say. Mention file types. Be more specific about scope.
Test MCP independently first: Ask Claude to call the MCP tool directly (without the skill). If that fails, the issue is the MCP connection — not your skill instructions.
Add negative triggers in description: "Do NOT use for simple data exploration (use data-viz skill instead)."
Be more specific about scope: "PayFlow payment processing for e-commerce. Not for general financial queries."
Instructions too verbose — move detail to references/. Put critical rules at the top. Keep SKILL.md under 5,000 words.
For critical validations: use a script in scripts/ rather than language instructions — code is deterministic, language interpretation is not.
Use the skill-creator skill built into Claude.ai. Say: "Use the skill-creator skill to help me build a skill for [your use case]." If you have an MCP server and know your top 2–3 workflows, you can have a working, tested skill in a single sitting.
Agentic CLI in your terminal. Full codebase context, file editing, bash execution, git automation, MCP, subagents, hooks, and skills. Runs in terminal, VS Code, JetBrains, or at claude.ai/code.
| Term | Definition |
|---|---|
| Session | One run of claude in a directory — from launch to exit |
| Context window | Total text held at once. Use /compact at >80%, /clear when switching tasks |
| CLAUDE.md | Markdown auto-loaded every session. Include: tech stack, coding rules, naming conventions, testing process, commands |
| Plan mode | Read-only — Claude proposes, does NOT edit files. Toggle Shift+Tab or /plan. Use before every large refactor |
| Accept Edits mode | Auto-approves all file edits for the session. Toggle Shift+Tab. Use only when you trust the direction |
| Subagent | Secondary Claude instance for a focused subtask in isolation — prevents context pollution. Defined in .claude/agents/ |
| Hook | Shell command at a defined event (PostToolUse, PreToolUse). Auto-runs tests, linting, or blocks unsafe commands |
| Skill | Packaged workflow in .claude/skills/name/SKILL.md. Skills win over legacy .claude/commands/ on name collision |
| Command | Description |
|---|---|
/clear | Wipe history, fresh start (alias: /reset, /new) |
/compact | Summarize conversation, free context (>80% rule) |
/context | Visualize what is loaded in context window |
/memory | View or edit CLAUDE.md files in scope |
/btw | Side question without polluting main conversation |
/rewind | Roll back conversation and code to checkpoint |
/add-dir | Grant Claude access to an additional directory |
| Command | Description |
|---|---|
/plan | Enter read-only plan mode — propose, don't edit |
/diff | Interactive viewer of all uncommitted changes |
/review | Code quality review |
/security-review | Security audit before committing changes |
/debug | Structured debugging playbook |
/mcp | Manage MCP server connections mid-session |
/agents | List, create, or edit subagents |
/hooks | Configure event hooks interactively |
/skills | List available skills |
| Command | Description |
|---|---|
/help | List every available command |
/model | View or switch model (/model sonnet) |
/effort low|max | Adjust reasoning depth and token spend |
/cost | Show token usage and session cost |
/config | Open settings — theme, defaults |
/permissions | View/change which tools require approval |
/doctor | Diagnose install and environment issues |
/init | Generate CLAUDE.md for a new project |
/install-github-app | Enable automatic PR review in GitHub Actions |
/ide | Connect to VS Code or JetBrains inline |
| Key | Action |
|---|---|
| Shift+Tab | Cycle: default → acceptEdits → plan |
| Esc | Interrupt Claude mid-response |
| Esc Esc | Open rewind / checkpoint menu |
| Ctrl+C | Cancel input or exit on empty prompt |
| Ctrl+R | Reverse search prompt history |
| @ + path | Reference a file/directory in prompt |
| / | Open the command menu |
# Pipe data into Claude
git diff main | claude -p "Review for security issues and breaking changes"
cat error.log | claude -p "Find the root cause"
# Structured JSON output for scripts and pipelines
claude -p "Analyze this PR" --output-format json
# Streaming JSON for real-time processing
claude -p "Generate report" --output-format stream-json
# One-shot prompt and exit
claude -p "Quick fix for this bug" --max-turns 1
# Resume last session
claude --resume # alias: claude -c
# Skip permission prompts (trusted/isolated environments only)
claude --dangerously-skip-permissionsMost developers use Claude Code only for snippets and bug fixes. Power users build repeatable systems around it. These 35 tricks unlock the full platform — applying even 5 will measurably improve your workflow.
Clears old context so Claude does not mix unrelated details. The single most important context hygiene habit for clean results.
Summarizes heavy conversations while preserving development direction. Use when context exceeds 80% — preserves key decisions.
Reopens the previous session to continue without repeating project context. Alias: claude -c. Pairs with CLAUDE.md for full project continuity.
Lets Claude edit files and run commands without approval prompts. Use only in isolated or trusted environments — never production.
Adjusts reasoning depth. Use low for quick fixes and max for architecture decisions and security reviews. Controls token cost directly.
Gives Claude your project structure, coding rules, commands, and preferences. Include: tech stack, commands, testing process, folder structure, coding rules, naming conventions.
Stores personal coding preferences so Claude follows them across ALL projects. Your persistent engineering identity across every codebase.
Tell Claude what to avoid: "Don't use Axios, use fetch. Avoid class components, use hooks only." Reduces repeated mistakes significantly across sessions.
Use separate git worktrees to run parallel Claude sessions: one fixing a bug, another reviewing code, another building a feature — simultaneously without conflicts.
Creates an isolated branch folder for conflict-free parallel work:git worktree add ../myapp-review -b review/cleanup
Auto-runs tests, linting, or quality checks after Claude edits files. Quality gates that fire without you asking — reliably, every time.
Blocks risky commands before Claude runs destructive or unsafe actions. Your safety net. Use scripts here — code is deterministic, language is not.
Configures workflow hooks interactively without manually editing JSON settings files. Hooks make Claude safer and more trustworthy during development.
Creates focused agents for reviews, security checks, documentation, or frontend work. Each runs in isolation, preventing context pollution of the main session.
Restricts each agent to relevant project areas only. Archetypes: Code Reviewer, Security Scanner, Docs Writer, Frontend Helper, DevOps Assistant.
Opus: Architecture, deep reasoning, security decisions.
Sonnet: Implementation, refactoring, feature work.
Haiku: Quick fixes, tests, repetitive tasks. Smart routing saves cost without sacrificing quality where it matters.
Enables Claude to review pull requests automatically inside GitHub Actions workflows — no manual trigger needed per PR.
Diagnoses setup, installation, and configuration issues without manual guesswork. Run this first whenever anything behaves unexpectedly.
Shows token usage and session cost. Combine with /effort low and /model haiku to control spend on large or repetitive projects.
Connects Claude Code directly with VS Code or JetBrains — see diffs, edits, and output inline in your editor without switching to terminal.
Send logs, diffs, errors, or command outputs directly into Claude:cat error.log | claude -p "find the root cause"
Returns structured responses parseable by scripts, dashboards, and automation pipelines. No regex — clean JSON every time.
Streams Claude's output live for real-time processing. Ideal for CI/CD display pipelines and progress dashboards in automation.
One-shot prompt and exits after one answer. No interactive back-and-forth. Perfect for scripted automation, cron jobs, and CI/CD steps.
Turns repeated prompts into reusable slash commands: /review, /deploy, /document, /commit. Note: .claude/skills/ is now preferred — Skills win on name collision.
Injects live git status, diffs, logs, or test results automatically into command context before Claude sees it — dynamic, no manual pasting.
Limits what each command can access. Makes workflows safer and controlled by design. Pair with PreToolUse hooks for maximum safety.
Define expected behavior before Claude writes any implementation code. Catches logic gaps, edge cases, and security issues before they compound.
Use /plan to have Claude propose an approach, discuss it with you, then execute only after your explicit approval. Fewer wrong changes on large refactors.
Catches logic gaps, bugs, edge cases, and security issues right after the build phase — while context is fresh. Use the code-review-agents skill.
Connects Claude with project-specific APIs, databases, and cloud tools via .mcp.json. Each project gets its own service connections automatically on session start.
Manages connected MCP servers without leaving your active session. Add, remove, or reconnect services without restarting Claude.
Opens rewind history to restore previous file states safely — your undo button for Claude's changes across file edits within the session.
Reference @src/handler.go in your prompt to let Claude read that file directly without pasting code — precise, zero bloat.
Gives Claude access to multiple folders, repositories, or monorepo sections beyond the current working directory — essential for large workspaces.
/init → review CLAUDE.md → /plan → discuss → Shift+Tab acceptEdits → execute → /diff → /security-review → commit. Run /compact between logical chunks. Check /cost periodically. Eliminates most surprises on large refactors.
Desktop-level agentic AI for knowledge workers — Claude Code without the terminal. Autonomous multi-step task execution with a visual approval flow. Zero coding required. Runs only in the desktop app (Mac/Windows). Supports scheduled overnight tasks via the /schedule plugin.
Point Coworker at a folder: "Organize these downloads into categories", "Draft summaries from these meeting notes", "Rename all files by date." Plans and executes autonomously.
Before any sensitive action (moving, renaming, deleting), Coworker shows a clear GUI plan. Approve or modify each step before anything irreversible executes. Full audit trail.
Connects Chrome, Asana, Notion, and other apps — pulling data from web applications and synthesizing it locally into reports, structured documents, and dashboards.
Use the /schedule plugin to set recurring Cowork tasks. Write your full prompt — Claude runs it before you wake up and saves a briefing doc to Claude Outputs. Claude works while you sleep.
Everything outside Claude Outputs is read-only. Name work files clearly: project_sent_v1.md. On Mac, create a Text Shortcut (e.g. /prompt) to expand your master template anywhere.
Instead of writing paragraphs of instructions, flip the script. Say: "We're getting sidetracked. Generate an AskUserQuest." Claude generates a structured multi-select questionnaire. You click answers in under a minute. Claude plans, you approve, it executes. This is what good AI-assisted planning feels like.
Browse plugins in the desktop app to add a sidebar matching your workflow. Types: Marketing (/marketing:draft-content), Data (CSV + dashboard explore), Legal (contract review in sidebar). Master one — then expand.
The most common mistake: using Claude Chat for tasks that Claude Code handles better, or Coworker for tasks that Chat resolves faster. Use this matrix to route correctly every time.
| Scenario | Best tool | Reasoning |
|---|---|---|
| Draft an email, brainstorm architecture, summarize a PDF | Claude Chat | Fast, visual, isolated from file system — no risk, instant output |
| Refactor a 10,000-line Go service and fix failing tests | Claude Code | Full codebase context, autonomous test execution, git integration |
| Extract data from 50 PDFs into an Excel dashboard | Coworker | File management, multi-step planning, zero coding required |
| Automate code review in a CI/CD pipeline | Claude Code | Non-interactive headless flags, fully pipeable, composable with bash |
| Organize 200 downloaded files into categorized folders | Coworker | Desktop-level file operations with visual approval flow and audit trail |
| Build an interactive React dashboard from scratch | Claude Chat | Artifacts render live in the browser panel — instant preview and iteration |
| Security audit on a PR before merging | Claude Code | /security-review + /diff provide git-aware structured analysis |
| Search Gmail, Calendar, and Notion in one workflow | Chat + MCP | MCP connectors natively available in the chat interface with one-click OAuth |
| Deep research across 15+ sources — competitive analysis | Claude Chat | Deep Research mode runs 10–20 searches and synthesizes a comprehensive report |
| Build a SKILL.md workflow package for your team | Claude Code | skill-creator skill, git-tracked, team-shareable via .claude/skills/ |
| Overnight report from Notion + Slack + Drive | Coworker | /schedule plugin runs multi-connector tasks overnight — briefing doc on wakeup |
| Generate a DOCX, PPTX, or XLSX file | Claude Chat | Built-in document skills (docx, pptx, xlsx) activate automatically via skill system |
Problem-first: "I need to set up a project workspace" → your skill orchestrates the right MCP calls. Users describe outcomes; the skill handles the tools. Tool-first: "I have Notion MCP connected" → your skill teaches Claude the optimal workflows. Users have access; the skill provides expertise. Knowing which framing fits your use case helps you choose the right skill pattern.
Based on the "How to Master Claude in One Week" guide. Follow these 10 steps in sequence to transition from basic chatting to advanced workflow automation with scheduled autonomous tasks.
/prompt) that expands it anywhere. Name work files clearly: project_sent_v1.md./schedule plugin. Write your Cowork prompt in full. Set it to run before you wake up. You open your laptop to a completed briefing doc. Claude works while you sleep.The best strategy shifts from manual prompt writing to systemic context alignment. These principles apply across all three products and compound over time with consistent use.
Lead every prompt with who Claude should be, then what it should do, then how to present output.
Wrap inputs with XML tags — Claude parses them reliably and produces outputs matching your schema exactly.
Use one master template that reads your about-me.md before every task. Set an OS-level text shortcut (/prompt on Mac) to expand it instantly anywhere. Guarantees context alignment without retyping. Pairs with the Cowork folder structure for maximum consistency.
Add "think through the tradeoffs first, then give your recommendation" — triggers deeper reasoning especially for architecture decisions and debugging. Particularly effective with Opus on complex security or system design tasks.
Claude holds full context across a conversation. "Now refactor that for production Go" or "add ClickHouse integration to the previous example" — no need to re-paste code. In Claude Code, use claude --resume to continue across terminal sessions.
Add only files necessary for the current task. Run /compact when context exceeds 80% and /clear when switching tasks entirely. Always review /diff before applying destructive changes. Bloated context = drift, higher cost, lower quality.
Anthropic's official free learning platform, launched March 2026. All 17 courses are self-paced, include video lessons and quizzes, and award an official completion certificate ready for LinkedIn. No Anthropic account required to enroll — sign up at anthropic.skilljar.com with any email.
Every course awards an official Anthropic completion certificate. These credentials carry weight in job applications — they prove structural understanding of Claude, not just casual familiarity. No paywall. No prerequisites for the non-technical track. Certificates are shareable directly on LinkedIn.
If you take only one course, take this one. Teaches everyday use: writing emails, organizing data, analyzing documents, drafting content. No technical background needed. The definitive first step — after this, you know which direction to go deeper.
Co-developed with academic partners around the "4E Framework": Effective, Efficient, Ethical, Safe. Teaches thinking frameworks — where AI capability boundaries lie, which tasks to delegate, and when human judgment cannot be replaced. Taught by Prof. Joseph Feller (UCC) and Prof. Rick Dakan (Ringling College).
Academic research, career planning, and learning strategies using Claude. Practical guidance for responsible AI use in coursework, professional preparation, and academic integrity. Designed for undergrad and graduate learners.
Integrating AI into instruction, curriculum design, and assessment. For teachers and professors who want to guide students on responsible AI use — covers classroom strategies, assignment design, and academic integrity frameworks.
How to teach others to use AI effectively. Covers scenario design, learning outcome evaluation, and instructional frameworks. For L&D professionals, corporate trainers, and instructional designers building internal AI programs.
Co-developed with GivingTuesday. Practical AI adoption for mission-driven organizations with limited technical staff and constrained budgets. Covers fundraising, grant writing, program management, and volunteer coordination use cases.
The first official course on Anthropic's agentic desktop agent. Designed for non-technical users — covers task delegation, approval flows, file automation, connector workflows, the 4-folder workspace structure, and scheduled task setup with the /schedule plugin.
How a generative model actually works — real capabilities, hallucination patterns, reasoning limits, context window mechanics, and operational boundaries. Recommended for all roles before going deeper into any technical track. Essential for setting correct expectations.
First API call through to production patterns — tool use, function calling, streaming, batch processing, prompt caching. Essential starting point for every Python developer building on Claude. Covers the full Messages API, current model strings, and SDK usage.
Flagship developer course. Installation through advanced integration: context management, CLAUDE.md setup, MCP server configuration, tool chaining, slash commands, CI/CD scripting, hooks, subagents, and custom skills. Official Anthropic certificate — a meaningful differentiator in 2026 hiring.
Design, write, and compose Claude Skills — reusable SKILL.md workflow packages with progressive disclosure loading. Covers auto-invocation rules, skill composability, YAML frontmatter, the 5 workflow patterns, the skill-creator skill, and Skills 2.0 executable format from Q1 2026.
Build your first MCP server from scratch — expose tools, resources, and prompts to Claude. Full protocol spec with Python implementations, OAuth-protected endpoints, and real integration examples. Covers .mcp.json per-project configuration and Claude Code integration.
Async MCP patterns, OAuth-protected servers, dynamic client registration, multi-server orchestration (Multi-MCP coordination pattern), production deployment at scale. For builders moving MCP integrations from prototype to production.
Delegate tasks to isolated sub-agents — context isolation, coordination patterns, agent teams, and git worktree isolation for parallel development. For complex agentic architectures requiring parallel execution and specialized task handling via .claude/agents/.
Deploy Claude via AWS — Bedrock API, IAM configuration, model invocation, knowledge bases, and production patterns for enterprise AWS environments. Covers Bedrock Agents and guardrails for regulated industry deployments.
Deploy Claude via GCP — Vertex AI integration, service accounts, regional endpoints, and production-grade deployment patterns for Google Cloud environments. Covers Vertex AI Extensions and enterprise security controls.
Anthropic's first professional certification program. Covers Claude Code, Skills architecture, Cowork, agent automation patterns, MCP server design, and production deployment. Includes a 12-week study plan, 25 practice questions, and exam prep guide. Targeted at engineers and architects building production Claude systems.
| Resource | URL | Best for |
|---|---|---|
| Anthropic Academy | anthropic.skilljar.com | All 17 free courses with official completion certificates |
| API documentation | docs.claude.com | Full API reference, models, tool use, streaming, prompt caching |
| Claude Code overview | docs.claude.com/claude-code | CLI install, slash commands, MCP, CLAUDE.md, skills, hooks |
| Prompt engineering guide | docs.claude.com/prompt-engineering | XML tags, chain-of-thought, system prompts, tool use patterns |
| MCP protocol spec | modelcontextprotocol.io | Build custom MCP servers, OAuth, connector patterns |
| Support center | support.claude.com | Plans, billing, feature limits, account issues |
| Claude Certifications | claudecertifications.com | Certified Architect program, exam prep, 25 practice questions |
| Anthropic news | anthropic.com/news | Model launches, feature announcements, research papers |
| Official skills repo | github.com/anthropics/claude-code | Official Anthropic skills — browse, fork, and customize |
| Class Central | classcentral.com/subject/claude | 600+ community courses on Claude across all platforms |