Study guides: MCP, Agent SDK & Skills

Learning to actually build AI agents —MCP, the Agent SDK, Skills— means crossing scattered official docs, specs that change between versions, and many details you only learn by tripping over them. There was no consolidated, example-driven study path in Spanish.

I wrote three independent study guides, verified against official documentation (MCP spec 2025-11-25 and Anthropic docs, audited as of 2026-05-15), with fictitious examples, exercises, glossaries, footguns and a 4-week learning path. They form a sequence: Agent SDK → Skills → MCP. Study material, not official docs — the official source always prevails.

Free, downloadable material so anyone with intermediate Python can learn to build with MCP, the Agent SDK and Skills from a single place. All examples are fictitious: zero private data.

• MCP — Study guide (122 pp)The Model Context Protocol end to end: host/client/server architecture, primitives (tools, resources, prompts, sampling), transports (stdio and Streamable HTTP), OAuth 2.1 authorization, building servers and clients in Python, security, debugging and 13 footguns. Against the 2025-11-25 spec.
• Agent Skills — The complete manual (80 pp)The layer that turns agents into specialists: what a skill is, anatomy and YAML frontmatter, how each surface discovers and loads them, authoring step by step, skills with code and with reference files, integration in Agent SDK / Claude Code / API, plugins and marketplace, best practices and security. 11 exercises.
• Claude Agent SDK in Python (53 pp)Why Anthropic externalized the Claude Code engine as a library: standard SDK (LLM as responder) vs Agent SDK (the loop lives inside the SDK), anatomy of an agent, 5 design patterns, cost and performance, and when NOT to migrate (over-engineering).