Explore Claude & AI concepts
Plain-language definitions of the terms that come up most in Claude certification exams — from tokens and tool use to agents, Claude Code and MCP.
Showing 41 of 41 concepts
AI foundations
Large language model (LLM)
A neural network trained on large amounts of text to predict the next token. That simple objective lets it write, summarise, reason through problems and use tools. Claude is a family of LLMs built by Anthropic.
Token
The unit a model reads and writes: a word, part of a word or a punctuation mark. Context limits, output limits and pricing are all measured in tokens.
Context window
The maximum number of tokens a model can consider at once, including the system prompt, the conversation, tool results and its own reply. Anything outside the window is invisible to the model.
Temperature
A setting that controls how random the output is. Lower values make answers more consistent; higher values make them more varied.
Hallucination
When a model states something false with confidence, such as an invented citation or API parameter. Grounding answers in supplied sources and allowing the model to say "I don’t know" both reduce it.
Embeddings
Numeric vectors that represent the meaning of text, so similar passages sit close together. They power semantic search and retrieval.
Model selection
Choosing a model tier by trading off capability, speed and cost. Larger models handle complex reasoning; smaller ones suit high-volume, latency-sensitive tasks such as routing or classification.
Prompting
System prompt
Instructions supplied separately from the user’s messages that set the model’s role, rules, tone and context for the whole conversation.
Few-shot prompting
Including a handful of worked examples in the prompt so the model copies their format and approach. Varied examples work better than near-identical ones.
Chain of thought
Asking the model to reason step by step before answering. It improves accuracy on multi-step problems at the cost of more output tokens.
Extended thinking
A Claude feature that gives the model a budget of tokens to reason internally before it writes its final answer. Useful for hard analysis, maths and planning.
Claude API
Messages API
The core Claude API endpoint. You send a list of user and assistant messages (plus an optional system prompt and tools) and receive the model’s next message.
Tool use
Letting Claude call functions you define. You describe each tool with a name, description and JSON input schema; Claude returns a tool call, your code runs it and sends back the result.
Structured output
Getting responses in a fixed, machine-readable shape, usually JSON that matches a schema. Tool schemas or structured-output features are more reliable than asking for JSON in plain text.
Streaming
Receiving the response token by token as it is generated, so users see text immediately instead of waiting for the full reply.
Prompt caching
Reusing a long, unchanging prompt prefix, such as a system prompt or reference document, across requests. It cuts cost and latency for repeated calls.
Message batches
Submitting many requests to be processed asynchronously at lower cost. Good for large offline jobs where you don’t need an immediate answer.
Vision
Claude’s ability to read images and PDFs, such as charts, screenshots, scanned forms and diagrams, alongside text.
Agents
Agent
A system where the model decides its own next steps in a loop: it plans, calls tools, reads the results and continues until the task is done or a limit is reached.
Workflow vs agent
A workflow follows steps you define in code; an agent chooses its own path. Workflows are cheaper and more predictable, so use an agent only when the steps can’t be known in advance.
Agent loop
The cycle of model call, tool call and tool result that repeats until the model stops asking for tools. Production loops need stop conditions, step budgets and error handling.
Orchestrator and subagents
A pattern where a lead agent breaks a task into parts and hands each to a subagent with its own focused context, then combines the results.
Human in the loop
Requiring a person to approve high-impact actions, such as sending money, deleting data or emailing customers, before the agent performs them.
Claude Agent SDK
Anthropic’s library for building agents on the same foundations as Claude Code, including the agent loop, tools, permissions and context management.
Claude Code
Claude Code
Anthropic’s agentic coding tool. It reads your codebase, edits files, runs commands and works through tasks from the terminal, IDE, desktop app or CI.
CLAUDE.md
A memory file Claude Code loads automatically with project conventions, commands and rules. It can live at user, project or directory level.
Hooks
Shell commands that run automatically at set points, such as before or after a tool call. They enforce rules deterministically instead of relying on the model to remember them.
Slash commands
Reusable prompts saved as commands a team can run, for example to review a pull request or write release notes in a house style.
Permissions
Settings that control which tools, commands and files Claude Code may use without asking. Tight permissions are the first line of defence in automated setups.
Headless mode
Running Claude Code non-interactively from a script or CI pipeline, with a prompt in and results out.
MCP
Model Context Protocol (MCP)
An open standard for connecting AI applications to external tools and data. A service wraps itself once as an MCP server, and any MCP-compatible client can use it.
MCP server and client
The server exposes capabilities, such as a database, ticket system or file store. The client lives inside the AI application and connects the model to those servers.
Tools, resources and prompts
The three main things an MCP server can offer. Tools are actions the model can call, resources are data it can read, and prompts are reusable templates a user can pick.
Tool description
The name and text that tell the model what a tool does and when to use it. Clear descriptions and precise input schemas matter as much as the tool’s code.
Safety & reliability
Prompt injection
An attack where instructions hidden in content the model reads, such as a web page, email or tool result, try to override its real instructions.
Retrieval-augmented generation (RAG)
Searching a knowledge source for relevant passages and adding them to the prompt so answers are grounded in current, specific information.
Context compaction
Summarising or trimming older parts of a long conversation so an agent can keep working without running out of context or losing key facts.
Evaluations (evals)
Test sets that measure how well a prompt, model or agent performs on real examples. Evals turn prompt changes from guesswork into measurable improvements.
Guardrails
Deterministic checks around a model, such as input filters, output validation, rate limits and approval steps, that keep a system safe even when the model makes a mistake.
Constitutional AI
Anthropic’s training approach in which a model learns to follow a written set of principles, used to make Claude helpful, honest and harmless.
Check the spelling, try a broader term, or search all topics.
Turn definitions into exam readiness
Knowing a term is the first step; the exams test when to use it. Our study guide connects these concepts to each exam domain, and the study materials go deeper on each one. Practice sets are available for Associate, Developer, Architect — Foundations and Architect — Professional.
Test what you know
Scenario questions put these concepts to work. Try a timed mock exam and see which ones need another look.
ClaudeMock is an independent practice-exam provider and is not affiliated with or endorsed by Anthropic. Claude and Anthropic are trademarks of Anthropic PBC. Definitions are simplified for study purposes; see Anthropic’s documentation for full technical detail.