Glossary
50+ terms

A

Agent SDK

Anthropic's official SDK for building agents on top of Claude. It standardizes the tool-use loop, memory, subagents, and permissions. Claude Code itself is built on the Agent SDK internally. More →

agentskills.io

The community registry for Claude Skills. Users publish, search, and share SKILL.md-based procedure packs. Acts as the extension hub around the official ecosystem. More →

Anthropic Labs

Where Anthropic ships experimental features to a limited audience first. Research Preview items — Routines and the early Claude Cowork, for example — land here before anywhere else. More →

APPI (Act on the Protection of Personal Information)

Japan's personal-information protection law (個人情報保護法). When using Claude at a Japanese company, cross-border transfer, consent, and safeguards all have to be considered under APPI. More →

Artifacts

A Claude feature that surfaces generated code, documents, and diagrams in a panel separate from the chat, where you can view and edit them. Introduced on Claude.ai in 2024. Became the foundation for Cowork. More →

Audit Log

The enterprise audit-log feature. Records member actions, API key issuance, and data access. Used as evidence for SOC 2 and ISO 27001 compliance. More →

B

BAA (Business Associate Agreement)

A partnership agreement under US HIPAA. Required between Anthropic and any customer handling PHI (protected health information). Available on the Enterprise plan, and relevant to Japanese healthcare companies with US operations or overseas expansion. More →

Bedrock

AWS's managed LLM service. Claude Opus, Sonnet, and Haiku all run on it. The AWS Tokyo region gives you a BYOK path with Japanese data residency. More →

Batch API

An Anthropic API feature for running large volumes of prompts asynchronously. Half the normal price. Good fit for large-scale evaluation or batch analysis where real-time isn't needed. More →

C

Claude Code (CLI)

Anthropic's official coding agent that runs in the terminal. Launches with the claude command. Has Plan Mode, Parallel Sessions, MCP — all of it. Aimed at developers. More →

Claude Code Desktop

The Code tab inside the Claude Desktop app. Major redesign went GA on 2026-04-14. Visual Diff, integrated terminal, drag-and-drop layout, and other GUI-only features. More →

Claude Design

Claude's specialist surface for turning Figma and UI mockups into production-quality HTML, CSS, and React. Automates the handoff between design and engineering. More →

Claude Desktop

The native desktop app for macOS and Windows. Holds Chat, Cowork, and Code as tabs. Also serves as an MCP host. More →

Claude Cowork

The surface where Claude co-edits documents, spreadsheets, and slides with a human. Evolved from Artifacts. Built for document-centric work. More →

Claude for Word

The official Microsoft Word add-in for calling Claude. Contract review, meeting-note cleanup, translation — all without leaving Word. More →

Cloudflare Pages

A static-site hosting service. This workshop site runs on Cloudflare Pages, deployed automatically from a GitHub push. More →

Computer Use

Claude's ability to see the screen and operate mouse and keyboard. Shipped in beta in October 2024. By 2026, OSWorld benchmark scores have climbed substantially. More →

Context Window

The maximum number of tokens the model can handle at once. Claude Opus 4.7 runs 200K standard with 1M tokens as an extended option — enough to take in a whole codebase or a long contract. More →

D

DocuSign (connector)

The official DocuSign connector for Enterprise customers. Lets Claude search, analyze, and check send-status on contracts. More →

DPA

Data Processing Agreement. Required between you and Anthropic (as data processor) for GDPR and APPI compliance. Standard on the Enterprise plan. More →

E

Enterprise plan

Anthropic's top-tier business plan. Includes SSO (SAML), SCIM, audit logs, ZDR, BAA, DPA, and dedicated support. Seat-based, starting at 100 seats. More →

F

FactSet (connector)

The official Claude connector for the FactSet financial data provider. Earnings, market data, and company financials become searchable from Claude. More →

G

GitHub Secret Scanning

GitHub's secret-detection service. Anthropic API key patterns are registered. If a key gets committed to a public repo, GitHub pings Anthropic and the key is revoked automatically. More →

H

Haiku 4.5

The fast, cheap member of the Claude family. Good at classification, summary, and extraction, plus high-volume batch work. Roughly a fifth of Opus 4.7's cost. More →

HIPAA

US Health Insurance Portability and Accountability Act. With the Enterprise plan plus a signed BAA, Claude can be used on PHI. More →

Hot reload (skills)

Edit a SKILL.md file in Claude Skills and the change takes effect immediately, no restart. Shortens the iteration loop when you're authoring skills. More →

I

ISO 27001

The international standard for information security management systems (ISMS). Anthropic holds the certification and ships the report in enterprise procurement. More →

J

JetBrains plugin (Claude Code)

The official Claude Code plugin for IntelliJ, PyCharm, WebStorm, and the rest of the JetBrains family. Same capabilities as the VS Code extension. More →

Japan residency

Data processing and storage inside Japan. Not available on a direct Anthropic contract today, but you can get there via AWS Bedrock or GCP Vertex AI Tokyo regions. More →

M

Max plan

The upper-tier individual plan (above Pro). Usage caps are much higher — built for power users who hit Opus frequently. More →

MCP (Model Context Protocol)

An open protocol that standardizes how Claude (and other LLMs) connect to external tools and data sources. Anthropic announced it in late 2024. By 2026 there are hundreds of server implementations. More →

MCP Server

A server process that speaks the MCP protocol. Official and unofficial servers exist for Slack, GitHub, Google Drive, Notion, and most major SaaS and data sources. More →

MFA

Multi-Factor Authentication. Claude accounts support TOTP, WebAuthn, and passkeys. In Enterprise setups, MFA is usually enforced at the SSO layer. More →

Microsoft Copilot Cowork

Microsoft's collaborative editing feature, occupying the same slot as Claude Cowork. Integrated into Word, Excel, and PowerPoint. Worth tracking as a direct competitor. More →

O

OAuth

The industry-standard protocol for delegating API access to a third-party app. Used widely in Claude's MCP connectors and Desktop Extensions for authentication. More →

OIDC

OpenID Connect. An authentication layer built on top of OAuth 2.0. One of two protocol choices for Enterprise SSO, alongside SAML. More →

Opus 4.7

The Claude flagship released on 2026-04-16. Tuned for long, complex agent runs, large-scale coding, and strategic reasoning. Available from Claude Code v2.1.111 onward. More →

OSWorld benchmark

Measures how well a model performs real tasks on real OS screens and applications. The industry go-to for evaluating Computer Use and agent capabilities. More →

P

Parallel Session

A Claude Code feature for running multiple sessions at the same time. Different repos, different branches, different tasks, all in parallel. More →

Plan Mode

A Claude Code mode where the agent shows its plan and waits for approval before implementing. Toggle with Shift+Tab. Works as a safety valve for destructive changes. More →

Plugin (Claude Code)

The extension mechanism for Claude Code. Slash commands, agents, and hooks can be packaged and distributed as a single plugin. More →

Prompt Caching

An Anthropic API feature that caches long system prompts or large context so subsequent calls hit the cache. Input token cost drops by up to 90% on reuse. More →

Pro plan

The standard paid individual plan. Higher usage than Free, a choice between Opus and Sonnet, and Projects. More →

R

RBAC

Role-Based Access Control. Standard model for assigning permissions to roles like Owner, Admin, and Member. Available on the Enterprise plan. More →

Research Preview

Anthropic's pre-GA release stage. Routines, Computer Use, and others appeared here first. No SLA, no stability promises. More →

Routines (Claude Code)

Cloud automation for Claude Code, announced on 2026-04-14. Run agents on a cron-like schedule or trigger them from events. More →

S

SAML

Security Assertion Markup Language. The standard protocol for Enterprise SSO, connecting Claude to IdPs like Okta, Azure AD, or Google Workspace. More →

SCIM

System for Cross-domain Identity Management. Standard protocol for automatic user provisioning and deprovisioning from an IdP. Enterprise-only. More →

Side Chat

A Claude Code Desktop feature. ⌘+; or Ctrl+; opens a side thread where you can ask a question without derailing the main one. More →

Skills

Reusable procedure packs based on SKILL.md. Claude Code and Claude Desktop invoke them as /skill-name, running specialist procedures for a specific domain. More →

SKILL.md

The core Markdown file for a Skill. Front matter defines name, trigger, and permissions. The body holds the procedure. Hot-reloads on edit. More →

Slash Command

Claude Code and Desktop commands starting with /. Built-in or user-defined — /plan, /review, /ultrareview, and so on. Extensible through plugins. More →

SOC 2 Type II

A US report that validates security and availability through periodic audit. Anthropic holds a SOC 2 Type II report, shared under NDA during enterprise procurement. More →

Sonnet 4.6

The balanced member of the Claude family. Faster and cheaper than Opus. Used as the main tier for code editing and medium-scale document work. More →

SSO

Single Sign-On. On the Enterprise plan, supported via SAML or OIDC, connecting to your IdP and centralizing employee account management. More →

Subagent

A specialist sub-agent the main agent can call out to. Claude Code and the Agent SDK both support them. Used for search, review, testing — tasks that benefit from isolation so they don't pollute the main context. More →

T

Team plan

The business plan aimed at small to mid-sized teams. Limited SSO and centralized billing. Minimum 5 seats. Often chosen as a stepping stone before Enterprise. More →

TLS

Transport Layer Security. Traffic to the Anthropic API, Claude Desktop, and MCP servers is encrypted with TLS 1.2 or higher. A baseline item on any security audit. More →

U

Usage limit

Per-plan caps on messages, tokens, and time windows. Pro / Max / Team / Enterprise each raise the ceiling; Enterprise customers can negotiate custom limits. More →

V

Vertex AI

Google Cloud's managed LLM platform. Claude Opus, Sonnet, and Haiku all run on it. The Tokyo region gives you a BYOK path with Japanese data residency. More →

Visual Diff

A GUI-only feature in Claude Code Desktop. Compare large changes by file and by hunk. Leave inline comments. More →

W

Worktree (Git)

A Git feature for checking out different branches of the same repo into separate directories at the same time. Pairs well with Claude Code Parallel Sessions when you want multiple tasks running in parallel. More →

Z

ZDR (Zero Data Retention)

An Anthropic option: no training use, no log retention. Available on the Enterprise plan. Often a hard requirement for organizations handling sensitive data. More →