MCP Deep Dive.

Model Context Protocol. Anthropic's standard for AI agents to talk to external tools (databases, APIs, files). "USB-C for AI."

A process speaking the MCP protocol that exposes external resources (DB, API, files) as tools. Connects via stdio or HTTP.

A consumer that connects to one or more MCP servers. Claude Code, Cursor, Windsurf are all MCP clients.

A pre-baked prompt template offered by an MCP server. Surfaces as a slash command; the third pillar with tools and resources.

Content an MCP server exposes for reading (file, page, record). Optional context for the model.

When the model decides to invoke a function instead of generating text. Same as "function calling." The core of agentic behavior.

Designing what, how much, and in what order to show a model. The grown-up version of prompt engineering — the real lever in agentic systems.

The maximum number of tokens an LLM can "see" at once. When the window fills up, the model starts forgetting or the conversation gets compacted.

A modular capability bundle handed to an agent. Markdown instructions + tools + examples. Claude Code Skills productizes this.