#ai
#llm
#mcp
#education
How to Learn AI, LLM and MCP: An Annotated Resource Guide
Length:
10 min
Published:
June 10, 2026
Workshops and courses give you a foundation. What you do with it depends on what you study on your own.
This guide focuses on primary sources — official documentation, foundational papers, and materials from the people who built the technology. Third-party resources appear only where they explain things better or more clearly than the primary source.
How to read this guide:
- Official documentation and primary sources (Anthropic, MCP project, OWASP, NIST, original paper authors) form the backbone. Third parties are the exception.
- Most high-quality resources are in English — language is marked for each entry.
- The practical section focuses on Claude (Claude Code / Cowork), but the theory — LLM, prompting, security — is vendor-neutral.
- Level markers: 🟢 beginner · 🔵 intermediate · 🔴 advanced
Where to Start
No time for the full list? These eight resources cover the essentials: how LLMs work, how to write good prompts, and how to build your first agent.
- The Illustrated Transformer 🔵 EN or 3Blue1Brown video 🟢 EN — how LLMs work under the hood.
- Anthropic: Prompting best practices 🔵 EN — how to write prompts.
- Anthropic: Building Effective Agents 🔵 EN — what agents are (and aren't). The most-cited text on the topic.
- Claude Code: Best practices + Memory / CLAUDE.md 🔵 EN — how to manage agents in practice.
- What is MCP 🟢 EN — how agents connect to data and tools.
- OWASP Top 10 for LLM Applications (2025) + The lethal trifecta 🔵 EN — what to watch out for.
- Hands-on: Claude Code in Action or MCP: Build Rich-Context AI Apps 🔵 EN
- Elements of AI 🟢 EN — complete introduction to what AI is and isn't. Free, with certificate.
Generative AI and LLM Fundamentals
Understanding how LLMs work helps you write better prompts, configure agents, and reason about model behavior, especially failure modes.
How Transformers and LLMs Work
- Attention Is All You Need — Vaswani et al. (Google), 2017 · paper · 🔴 EN. The original paper that introduced the transformer architecture. The primary source; skimming is fine.
- The Illustrated Transformer — Jay Alammar · blog · 🔵 EN. The best visual explanation of self-attention and Q/K/V.
- Transformers, the tech behind LLMs — 3Blue1Brown · video (+ written version) · 🟢 EN. Visually excellent introduction — tokens, embeddings, attention, prediction.
Tokenization
- Understanding GPT tokenizers — Simon Willison · blog · 🟢 EN. Why " the" ≠ "The" and why non-English text costs more tokens.
- Tiktokenizer — interactive tool · 🟢 EN. See text break into tokens in real time. (Alt.: OpenAI Tokenizer.)
Embeddings
- Introduction to Embeddings — Cohere docs · 🔵 EN. Embeddings as vectors, similarity, the foundation of semantic search and RAG.
Context Windows
- What is a context window? — IBM Think · 🔵 EN. Context as working memory, the cost of attention, and the "lost in the middle" problem.
Sampling Parameters (temperature, top-p)
- What is LLM Temperature? — IBM Think · 🔵 EN. The best vendor-neutral explanation of temperature, top-k, and top-p.
- ⚠️ Note: the latest Claude models (Opus 4.7+) no longer accept
temperature/top_p, but the underlying theory remains valid across models.
- ⚠️ Note: the latest Claude models (Opus 4.7+) no longer accept
Model Selection
- Model selection guide — OpenAI · 🔵 EN. A clear framework: accuracy first, then cost and latency.
Prompt Engineering
Prompting is a discipline with well-documented techniques — not a guessing game. Anthropic's documentation is the best public reference.
- Prompt engineering — overview — Anthropic · docs · 🟢 EN.
- Claude: Prompting best practices — Anthropic · docs · 🔵 EN. The main continuously updated reference for prompting techniques.
- Interactive Prompt Engineering Tutorial — Anthropic · course/repo · 🟢🔵 EN. 9 chapters with exercises.
- Prompt Engineering Guide — DAIR.AI · 🟢🔴 EN. The most comprehensive open vendor-neutral reference: zero/few-shot, CoT, ReAct, RAG, agents.
- Best practices for prompt engineering (OpenAI API) — OpenAI · 🟢 EN.
Building AI Agents
Know what makes an agent an agent and what makes it badly designed. These texts are the foundation before you start building.
Core Concepts (Vendor-Neutral)
- Building Effective Agents — Anthropic Engineering · 🔵 EN. Workflows vs. agents, 5 patterns, "keep it simple". Required reading.
- Effective context engineering for AI agents — Anthropic Engineering · 🔴 EN. Context as a scarce resource, context rot, just-in-time retrieval, sub-agents. Pairs with our article Context Engineering: The Developer Skill That Decides AI Quality.
- Writing effective tools for agents — Anthropic Engineering · 🔴 EN.
Claude Code
- Claude Code — Overview · 🟢 EN.
- Best practices · 🔵 EN. explore → plan → code → commit, context hygiene, sub-agents.
- Memory & CLAUDE.md · 🔵 EN.
- Custom subagents · 🔵 EN.
- Hooks · 🔴 EN.
Claude Desktop and Cowork
- Claude Cowork — product · 🟢 EN. Cowork is now generally available on paid plans.
- Get started with Claude Cowork · 🟢 EN.
- Get started with the Desktop app · 🟢 EN.
Agent Skills
- Agent Skills — Overview · 🟢🔵 EN.
- Skill authoring best practices · 🔵 EN.
- Use Skills in Claude Code · 🔵 EN.
- Equipping agents for the real world with Agent Skills · 🔵 EN.
- anthropics/skills · 🔵🔴 EN. Reference open-source Skills.
- What are Skills? · 🟢 EN. Non-technical explanation.
MCP — Connecting Agents to Data and Tools
The Model Context Protocol is the standard that defines how agents communicate with external systems. For a deep analysis of risks, see our article MCP Under the Microscope: How AI Agents Talk to Tools and What Risks This Brings.
Official MCP Documentation
- What is MCP (Get started) — MCP project (Linux Foundation) · 🟢 EN. "The USB-C for AI".
- Architecture overview · 🔵 EN. Host/client/server, JSON-RPC, primitives (tools/resources/prompts).
- Build an MCP server · 🔵 EN.
- Specification (2025-11-25) · 🔴 EN. Current stable specification.
- GitHub: modelcontextprotocol · 🔵 EN. Ready-made servers in MCP Registry.
MCP in Claude and Microsoft
- Introducing the Model Context Protocol — Anthropic · 🟢 EN. Why MCP was built.
- Connect Claude Code to tools via MCP · 🔵 EN.
- Get started with custom connectors (remote MCP) · 🟢 EN. Note: connections originate from Anthropic's cloud IPs (relevant for enterprise network allowlisting).
- microsoft/mcp — official catalog · 🔵 EN. Azure, Microsoft 365 (Mail/Teams/SharePoint), Dataverse, SQL, Playwright.
- Microsoft Learn MCP Server · 🟢🔵 EN.
Vendor-Neutral Explanation
- What is the Model Context Protocol (MCP)? — Cloudflare · 🟢 EN. Including what MCP doesn't handle by itself (auth, encryption).
Security, Governance and Best Practices
Read these before giving an agent access to internal systems. These risks are asymmetric: the cost of understanding them is low; the cost of an incident is not. For a deeper look, see our article Agentic AI Security.
LLM and Agent Risks
- OWASP Top 10 for LLM Applications (2025) — OWASP GenAI Security Project · 🔵 EN. Risk vocabulary: prompt injection, data leakage, supply chain.
- OWASP — Agentic Security Initiative · 🔴 EN. Threats specific to agents: goal hijacking, tool abuse, memory poisoning.
Prompt Injection and the "Lethal Trifecta"
- The lethal trifecta for AI agents — Simon Willison · 🔵 EN. (1) access to private data + (2) untrusted content + (3) ability to communicate outward.
- Prompt injection — article series — Simon Willison · 🔵 EN.
- Google's Approach to AI Agent Security · 🔴 EN. Three principles: clear human owners, limited permissions, observable actions.
MCP and Claude Code Security
- MCP Authorization (specification) · 🔴 EN. OAuth 2.1, token binding.
- A Practical Guide for Secure MCP Server Development — OWASP · 🔴 EN. 8 security domains; tool poisoning, "rug pulls".
- Claude Code — Security · 🔵 EN. Read-only default, isolated context for web fetch.
- OWASP AI Agent Security Cheat Sheet · 🔵 EN. Least-privilege, human-in-the-loop, audit logs, data classification.
Governance Frameworks
- NIST AI Risk Management Framework · 🔵 EN. GOVERN/MAP/MEASURE/MANAGE + profile for generative AI and critical infrastructure.
- EU AI Act — overview — European Commission · 🔵 EN. AI in critical infrastructure = high-risk category under the Act.
Free Hands-On Courses
Want to put theory into practice? These courses are free and well-structured.
- Anthropic Academy (catalog) · 🟢🔴 EN. Claude 101, Claude Code, MCP, Skills, sub-agents, Cowork, AI Fluency.
- Claude Code in Action · 🔵 EN.
- Introduction to Claude Cowork · 🟢 EN.
- MCP: Build Rich-Context AI Apps with Anthropic — DeepLearning.AI × Anthropic · 🔵 EN.
- Hugging Face AI Agents Course · 🔵 EN.
- The Hugging Face LLM Course · 🔵 EN.
- Generative AI for Beginners — Microsoft · 🟢 EN.
Czech Resources
Quality Czech-language AI learning resources are rare. These stand out.
- Elements of AI — Czech version — prg.ai + ČVUT + UK · 🟢 CZ. Entry: course.elementsofai.com/cs.
- AI v kontextu — MFF UK (Rudolf Rosa et al.) · 🟢🔵 CZ.
- Artificial Intelligence at Charles University — UK · 🟢🔵 CZ. Terminology glossary, AI Act overview, workshops.
- MCP Under the Microscope — DX Heroes · 🟢🔵 EN (+ Czech version).
- Context Engineering: The Developer Skill That Decides AI Quality — DX Heroes · 🔵 EN (+ Czech version).
Want to stay one step ahead?
Don't miss our best insights. No spam, just practical analyses, invitations to exclusive events, and podcast summaries delivered straight to your inbox.