A complete written reference for the Anthropic interactive guide. Mirrors every page and content module in the app: the company, its products, models, Claude Cowork, Claude Code and all of its primitives (sub-agents, hooks, MCP, checkpoints, background tasks, memory, headless mode, extended thinking), plus plugins, skills, slash commands, and a curated resources index.

1. Anthropic — the company
2. Products & surfaces
3. Models
4. Claude Cowork
5. Claude Code
6. Plugins
7. Skills
8. Slash commands
9. Resources

Anthropic — AI safety and research company building reliable, interpretable, steerable AI.

• Founded: 2021
• Headquarters: San Francisco, CA

Mission. Anthropic's mission is to ensure that humanity safely navigates the transition through transformative AI. The company builds frontier AI systems and studies their safety properties so the technology benefits society.

Key links

• Company
• Research
• Responsible Scaling Policy
• Acceptable Use Policy

Anthropic ships Claude across consumer chat, a developer API, an agentic coding tool, the Agent SDK, the major hyperscalers, and a desktop agentic surface for knowledge work.

Tagline: Your everyday thinking partner for writing, learning, and analysis.

Claude.ai is the consumer chat product. It supports text + image input, file uploads, web search, Projects (persistent context for collections of work), Artifacts (interactive documents, code, and apps generated inline), and connectors to Google Drive, Gmail, Calendar, GitHub, and more.

Key features

• Conversations with Opus, Sonnet, and Haiku
• Artifacts — interactive code, docs, diagrams, and mini apps
• Projects — persistent knowledge & instructions
• File uploads (PDFs, spreadsheets, images, code)
• Web search and citations
• Connectors: Drive, Gmail, GitHub, Jira, etc.
• Voice mode (mobile)
• Computer use (research preview)

Plans

Links: Open Claude.ai · Pricing · Help Center

Tagline: REST API at api.anthropic.com for building with Claude models.

The Claude API gives programmatic access to every Claude model plus higher-level managed agents. Available APIs include Messages (chat completions), Message Batches (50% discount async), Token Counting, Models, Files, Embeddings (beta), and the Managed Agents API.

Key features

• Messages API — primary chat endpoint
• Message Batches API — 50% cost reduction, async
• Tool use / function calling
• Vision (image input)
• Extended thinking (Sonnet/Haiku) and adaptive thinking (Opus)
• Prompt caching — up to 90% cost savings
• Files API and PDF support
• Service tiers: Standard, Priority, Batch
• Citations, Computer Use (beta), Code Execution (beta)

Links: API Overview · Get Started · Console · Pricing

Tagline: Agentic coding in your terminal, IDE, and CI.

Claude Code is Anthropic's AI-powered coding assistant. It understands entire codebases, edits multiple files, runs commands, browses the web, executes long-horizon tasks, and integrates with GitHub, GitLab, VS Code, JetBrains, Bedrock, Vertex, and the Agent SDK.

Key features

• Terminal CLI + IDE extensions (VS Code, JetBrains)
• GitHub & GitLab integration (PR review, @claude mentions)
• Sub-agents and headless mode
• Plugins, Skills, Slash commands, Hooks, MCP
• Background tasks, checkpoints, planning mode
• Agent SDK for building your own agents

Links: Overview · Quickstart · What's New

See the full Claude Code section below.

Tagline: Run Claude on AWS Bedrock, Google Vertex AI, and Microsoft Foundry.

Claude models are first-class citizens on AWS Bedrock, GCP Vertex AI, and Microsoft Foundry — letting enterprises consume Claude under existing cloud contracts, data-residency, and governance.

Key features

• AWS Bedrock — Claude Opus/Sonnet/Haiku
• Google Cloud Vertex AI
• Microsoft Foundry
• Regional availability for data residency
• Native IAM, VPC, KMS integration

Links: Bedrock · Vertex AI · Microsoft Foundry

Tagline: Build production agents on the same runtime as Claude Code.

The Agent SDK (TypeScript and Python) exposes the orchestration loop powering Claude Code: tool execution, sub-agents, hooks, MCP servers, sandboxed bash, file edits, and streaming. Use it to build coding agents, research agents, customer-support agents, and more.

Key features

• Python and TypeScript SDKs
• Built-in tools: Bash, Read, Edit, Glob, Grep, WebFetch
• Sub-agent spawning and parallel execution
• MCP server hosting
• Streaming events & hooks

Links: Agent SDK Docs · GitHub

Tagline: Anthropic's agentic AI for knowledge work — beyond the chat box.

Claude Cowork brings the same agentic architecture that powers Claude Code into Claude Desktop, for non-coding knowledge work. Describe an outcome, step away, and come back to a finished deliverable: a formatted spreadsheet, a slide deck, an organized folder, a research synthesis. Cowork runs locally on your Mac or Windows machine, executes code in an isolated VM, edits real files in folders you grant it, and can run on a schedule.

Key features

• Agentic Tasks mode inside Claude Desktop
• Direct local file read/write in folders you connect
• Sub-agent coordination & parallel workstreams
• Excel with working formulas, PowerPoint, formatted docs
• Scheduled & recurring tasks via /schedule
• Projects: persistent workspace + memory + instructions
• Mobile assignment (Pro/Max) — message from phone, Claude works on desktop
• Plugins: skills + connectors + sub-agents in one bundle
• Isolated VM for shell/code, layered permissions, deletion protection

Links: Cowork product page · Get started (Help Center) · Download Claude Desktop

See the dedicated Claude Cowork section for full coverage.

Three frontier tiers, all with 200K-token context (Sonnet has a 1M-token beta).

Most capable generally available model. Step-change improvement in agentic coding over Opus 4.6. Best for complex reasoning, multi-hour autonomous coding, and research.

Best combination of speed and intelligence. The default workhorse for most production workloads.

Fastest model with near-frontier intelligence. Ideal for high-volume classification, extraction, and latency-sensitive UX.

Anthropic's agentic AI for knowledge work — beyond the chat box.

Claude Cowork brings Claude Code's agentic architecture into Claude Desktop for non-coding work. You describe an outcome — a polished spreadsheet, a slide deck, an organized folder, a research synthesis — and Claude plans, breaks the work into subtasks, runs code in an isolated VM on your machine, edits real files in folders you've granted access to, and returns a finished deliverable. Pair it with Projects for persistent workspaces, and /schedule for tasks that run on a cadence.

• Plans: Pro · Max · Team · Enterprise
• Platforms: Claude Desktop for macOS · Claude Desktop for Windows
• Heads up: Web and mobile clients can assign tasks (Pro/Max), but Cowork itself runs on your desktop.

1. Plan — Claude analyzes your request and proposes an approach you can review or steer.
2. Decompose — Complex work is broken into subtasks, with sub-agents spawned in parallel where useful.
3. Execute — Shell commands and code run inside an isolated virtual machine on your computer.
4. Edit your files — Claude reads & writes real files in the folders you've connected — no manual upload/download.
5. Deliver — Polished outputs land in your filesystem. Step away mid-task and come back to results.

File & document management

• Organize a Downloads folder by type and date
• Drop receipts in a folder, get back a formatted expense report
• Batch-rename files with consistent YYYY-MM-DD patterns

Research & analysis

• Synthesize web searches, papers, and notes into a coherent report
• Extract themes and action items from meeting transcripts
• Surface patterns across journals or research files

Document creation

• Build Excel models with working formulas and multi-tab structure
• Turn rough notes into a presentation
• Convert voice memos and scattered notes into polished reports

Data & analysis

• Outlier detection, cross-tabulation, time-series analysis
• Generate charts directly from your data
• Clean and transform messy datasets

• Code execution isolation — Shell & code Claude writes runs inside an isolated VM, separate from your OS.
• Controlled file access — Claude can only touch folders you've explicitly connected.
• Network egress permissions — Claude follows your egress settings (note: web search & MCPs can bypass — admins can disable them).
• Deletion protection — Permanent deletes require an explicit Allow prompt.
• Local conversation history — Cowork stores history on your computer; not subject to Anthropic's retention timeframe.
• OpenTelemetry monitoring — Team & Enterprise admins can monitor Cowork activity via OTel (not a substitute for audit logs).

• Memory currently only persists inside Projects, not across standalone sessions
• Sessions and artifacts can't be shared with others
• Desktop app must remain open and the computer awake for tasks to run
• Cowork activity is not captured in Audit Logs, Compliance API, or Data Exports — don't use for regulated workloads
• Cowork consumes more usage than standard chat

• Cowork product page
• Get started guide
• Use Cowork safely
• Cowork for Team & Enterprise
• Schedule recurring tasks
• Assign tasks from anywhere
• Use plugins in Cowork
• Organize tasks with Projects
• Monitor with OpenTelemetry
• Download Claude Desktop

Claude Code is an agentic coding assistant that runs in your terminal, IDE, web, and CI. It reads and edits your codebase, executes commands, calls tools, and ships work end-to-end.

Install

npm install -g @anthropic-ai/claude-code

Spawn focused, isolated agents for parallel or specialized work. Sub-agents have their own context window, system prompt, and tool allowlist. Define them in .claude/agents/*.md with YAML frontmatter (name, description, tools). Claude delegates to them automatically based on the description, or you invoke explicitly with > use the code-reviewer subagent.

→ https://docs.claude.com/en/docs/claude-code/sub-agents

Shell commands that fire on lifecycle events. Hooks let you run scripts on PreToolUse, PostToolUse, UserPromptSubmit, Stop, SubagentStop, and Notification events. Great for auto-formatting, blocking dangerous commands, or sending desktop notifications.

→ https://docs.claude.com/en/docs/claude-code/hooks

Open protocol for connecting tools, data, and prompts to Claude. MCP is Anthropic's open standard for AI tool integration. Claude Code can connect to any MCP server (Postgres, Sentry, Linear, Figma, Playwright, Stripe, etc.) and expose your own server's tools, resources, and prompts.

→ https://modelcontextprotocol.io

Snapshot and rewind your working tree. Claude Code automatically checkpoints your repo before risky operations so you can /rewind to a known-good state.

→ https://docs.claude.com/en/docs/claude-code/checkpoints

Long-running jobs that survive your session. Kick off builds, migrations, or test suites in the background and let Claude monitor and react to results.

→ https://docs.claude.com/en/docs/claude-code/background-tasks

Persistent project, user, and team memory. CLAUDE.md files at the project, user (~/.claude/CLAUDE.md), and enterprise level provide persistent context — coding conventions, architecture notes, do's and don'ts.

→ https://docs.claude.com/en/docs/claude-code/memory

Scriptable, non-interactive execution. claude -p "refactor the auth module" runs non-interactively. Combine with --output-format json for piping into CI, cron, or other agents.

→ https://docs.claude.com/en/docs/claude-code/headless

Trigger deeper reasoning with magic words. Use think, think hard, think harder, or ultrathink in prompts to allocate more thinking budget on Sonnet/Haiku.

→ https://docs.claude.com/en/docs/claude-code/extended-thinking

Plugins package commands, sub-agents, hooks, and MCP servers into one shareable, installable unit.

• Slash commands (commands/*.md)
• Sub-agents (agents/*.md)
• Skills (skills/<name>/SKILL.md)
• Hooks (hooks/hooks.json)
• MCP servers (.mcp.json)
• A plugin.json manifest

• Standardize team workflows (review, deploy, release)
• Distribute domain expertise (security, accessibility, perf)
• Bundle MCP integrations with their helper commands
• Onboard new engineers with a single install

Links: Plugins docs · Plugin marketplaces

Skills are folders of instructions, scripts, and resources that teach Claude how to do specialized tasks. Claude auto-discovers them and loads only what it needs.

• Personal: ~/.claude/skills/<name>/
• Project: .claude/skills/<name>/
• Plugin-bundled: <plugin>/skills/<name>/
• Live change detection — edits hot-reload
• Nested directories scanned automatically

• pdf — read & generate PDFs
• docx — Word document authoring
• xlsx — spreadsheet creation
• pptx — slide decks
• artifacts-builder — produce Claude artifacts

---
name: brand-guidelines
description: Apply ACME brand voice, colors, and typography to any deliverable. Use when writing marketing copy or designing visuals.
---

→ https://docs.claude.com/en/docs/claude-code/skills

Slash commands are reusable prompts. Built-in commands control the session; custom commands live in .claude/commands/ (project) or ~/.claude/commands/ (personal).

---
allowed-tools: Bash(git:*), Read, Edit
argument-hint: <branch-name>
description: Create a release branch and bump version
---

Create a new release branch named `release/$1`, bump package.json,
and open a PR against main.

→ https://docs.claude.com/en/docs/claude-code/slash-commands

• Anthropic — Company
• Research
• Responsible Scaling Policy
• Acceptable Use Policy
• Trust Center

• Claude.ai
• Pricing
• Help Center

• API Overview
• Get Started
• Models Overview
• Console

• Claude Code Overview
• Quickstart
• Sub-agents
• Plugins
• Plugin Marketplaces
• Skills
• Slash Commands
• Hooks
• Memory (CLAUDE.md)
• Settings
• Agent SDK

• Model Context Protocol
• MCP Spec

• AWS Bedrock — Claude
• Vertex AI — Claude
• Microsoft Foundry

• Anthropic on GitHub
• Claude Code GitHub

• Claude Cowork — product page
• Get started with Claude Cowork
• Use Cowork safely
• Cowork for Team & Enterprise
• Schedule recurring tasks in Cowork
• Assign tasks from anywhere
• Use plugins in Cowork
• Organize tasks with Projects in Cowork
• Monitor Cowork with OpenTelemetry
• Future of AI at Work — Cowork webinar