Glossary for coding agents

Plain-English definitions of the words you hit when you work with Claude Code, Cursor and Codex: MCP, CLAUDE.md, the context window, prompt caching, token cost and more. Each entry is a short, practical explanation that links out to the guides and tools.

MCP (Model Context Protocol): MCP is an open standard that lets an AI agent call tools and read data outside the chat, such as a database, a repo or a task board.
MCP server: An MCP server is a small program that exposes a set of tools or data to any agent over the Model Context Protocol.
CLAUDE.md: CLAUDE.md is a file in your project root that Claude Code reads at the start of every session as its rules and context.
AGENTS.md: AGENTS.md is the project rules file that Codex and Cursor read, and that Claude Code reads too when no CLAUDE.md sits above it.
Context window: The context window is the amount of text, measured in tokens, a model can take in and attend to at once.
Prompt caching: Prompt caching reuses a processed part of your prompt across requests, so you pay a fraction of the input price for it instead of reprocessing it.
Cache read tokens: Cache read tokens are input tokens served from the prompt cache on a later request, billed at a fraction of the normal input price.
Cache write tokens: Cache write tokens are the tokens charged the first time content is stored in the prompt cache, at a small premium over the input price.
Agent session: An agent session is a single run of a coding agent: one continuous conversation whose memory resets when it ends.
Plan mode: Plan mode is a Claude Code mode where the agent researches and proposes a plan for your approval before it makes any edits.
Subagent: A subagent is a separate agent that a main agent spawns to handle a scoped task, returning only its result.
Hooks: Hooks are scripts the agent runtime runs automatically on events, such as before or after a tool call, to enforce rules or automate steps.
Slash commands: Slash commands are shortcuts, typed as /name, that run a packaged prompt or skill in a coding agent.
API-equivalent cost: API-equivalent cost is what a run’s tokens would cost at a model’s public API list prices, used as a comparable figure even when you pay a flat rate.
Acceptance criteria: Acceptance criteria are the specific, checkable conditions that have to be true for a task to count as done.