Claude Code Docs: A Map of What Is Where

The Claude Code docs are at code.claude.com/docs; the old docs.anthropic.com Claude Code pages now redirect there. The site is organized into Getting started, Build with Claude Code, Plugins, Administration, Configuration, Reference, Agent SDK and What’s New, with every page at code.claude.com/docs/en/<page>. For agents there is an index at /docs/llms.txt, a Markdown version of every page (add .md to the URL) and a documentation MCP server. Below is where to look for each common task, the pages worth reading that most people skip, and fenbs posts that go deeper.

How the site is organized

The left-hand navigation follows the order you tend to need things. Getting started covers the overview, the quickstart and core concepts. Build with Claude Code covers agents, MCP, skills, hooks and automation. Configuration holds settings, permissions, models and the interface. Reference holds the exhaustive tables: CLI flags, commands, environment variables, tools, hooks. Administration is for people rolling it out to an organization, and the Agent SDK section is for building your own agents on the same engine. As of September 30, 2026 the site is also translated into eleven other languages.

The map, by task

Getting started

Start with Overview (/en/overview) for the install command on each surface, then Quickstart (/en/quickstart) for a first real task. Advanced setup (/en/setup) has system requirements, other install methods, update channels and uninstalling. For a guided version, see how to install Claude Code and Claude Code for beginners.

Settings and permissions

The settings files and precedence↗ (opens in a new tab) page explains which of user, project, local and managed settings wins; All settings (/en/settings-reference) lists every key. For what Claude may do without asking, read Choose a permission mode (/en/permission-modes) and then the rules themselves.

The configure permissions↗ (opens in a new tab) page covers allow, ask and deny rules and their syntax, such as Bash(npm run test *); the sandboxing pages sit beside it. Auto-approve in Claude Code covers the modes in practice.

Memory and CLAUDE.md

How Claude remembers your project (/en/memory) covers CLAUDE.md at every level, .claude/rules/, imports, AGENTS.md and auto memory. Explore the .claude directory (/en/claude-directory) shows every file Claude Code reads and where. Depth: Claude Code memory and CLAUDE.md examples.

Hooks

There are two pages, and people often find only one. Automate actions with hooks (/en/hooks-guide) is the tutorial; Hooks reference (/en/hooks) has every event, input field and exit code. Worked examples: Claude Code hooks examples.

Skills and custom commands

Extend Claude with skills (/en/skills) covers the SKILL.md format, where skills live, the bundled skills and how custom slash commands fold into skills. Depth: how to create a Claude skill and Claude Code custom commands.

Subagents and parallel work

Create custom subagents (/en/sub-agents) has the file format and the built-in agents. The Agents and parallel work group next to it covers agent view, agent teams, worktrees and dynamic workflows. Depth: Claude Code subagent examples and running Claude Code tasks in parallel.

MCP

Connect to MCP servers (/en/mcp-quickstart) walks through adding one server end to end; Connect Claude Code to tools via MCP (/en/mcp) is the full reference for scopes, authentication and .mcp.json. When a server will not connect, see Claude Code MCP not working.

Plugins

Plugins have a whole section of their own: using them, creating them, running a marketplace, managing them for an organization, plus manifest and marketplace references. Start at Plugins overview (/en/plugins/overview). Depth: Claude Code plugins and plugins vs skills vs subagents.

Output styles and models

Output styles (/en/output-styles) and Model configuration (/en/model-config) sit under Configuration, with fast mode and the status line nearby. Depth: Claude Code output styles and Claude Code status line.

Headless runs and CI

The run Claude Code programmatically↗ (opens in a new tab) page covers claude -p, structured and streaming output, and permission settings for unattended runs. GitHub Actions (/en/github-actions) and GitLab CI/CD (/en/gitlab-ci-cd) are under Platforms and integrations. Depth: how to automate Claude Code tasks.

IDEs, desktop and web

Platforms and integrations (/en/platforms) compares every surface, then links out to VS Code, JetBrains, the desktop app, cloud sessions, mobile, Chrome and Slack. Depth: Claude Code in VS Code, Claude Code in JetBrains and Claude Code on the web.

Troubleshooting

Four pages, split by when the problem happens. Troubleshoot installation and login (/en/troubleshoot-install) opens with a table matching error text to a fix. Troubleshooting (/en/troubleshooting) covers high CPU or memory use, hangs and search problems, and Debug your configuration (/en/debug-your-config) helps when a setting, hook or rule is not taking effect.

The error reference↗ (opens in a new tab) explains each runtime error message: server errors, usage limits, authentication, network and request errors, and what Claude Code retries by itself. Depth: Claude Code not working.

The pages people miss

• How Claude Code works (/en/how-claude-code-works): the agentic loop, the tool categories, what Claude can access and how the context window fills. Short, and it explains most surprises.
• Extend Claude Code: a single decision page↗ (opens in a new tab) for when to use CLAUDE.md, skills, subagents, hooks, MCP or a plugin, and what each costs in context.
• Explore the context window (/en/context-window): an interactive simulation of what loads automatically and what each file read costs.
• Commands (/en/commands) and Environment variables (/en/env-vars): the complete lists, including the bundled skills you invoke like commands.
• Glossary (/en/glossary): the docs’ own definitions of agentic harness, session, surface and the rest.
• What’s new (/en/whats-new): a weekly digest, easier to read than the Changelog (/en/changelog), which lists every release.
• Best practices (/en/best-practices) and the Prompt library (/en/prompt-library): Anthropic’s own advice, with prompts you can copy.

Feeding the docs to an agent

Every page exists in two forms. The HTML page is for you; append .md to the URL and you get the page as Markdown, which is cheaper and cleaner for a model to read. Each Markdown page opens with a note pointing to the index, and the llms.txt index↗ (opens in a new tab) lists every page with a one-line description, grouped the same way as the navigation. A single llms-full.txt with the whole site also exists, but it runs to several megabytes, far too much to paste into a context window; fetch the index, then only the pages you need.

# The index: every page, one line each
curl -s https://code.claude.com/docs/llms.txt

# One page as Markdown
curl -s https://code.claude.com/docs/en/hooks-guide.md

# The hosted docs search as an MCP server (no sign-in needed)
claude mcp add --transport http claude-code-docs https://code.claude.com/docs/mcp

Inside Claude Code you often do not need any of that. Ask “how do I set up hooks?” and a built-in helper agent, claude-code-guide, looks up the documentation for you. It reaches code.claude.com to do so, so if your network blocks that host, only these documentation lookups stop working.

Read the docs against your version

The docs describe the latest release, and Claude Code ships often. Many sections say a behavior needs a given version or changed before one, so check yours with claude --version before assuming a page is wrong. If you are on the stable channel, you run a release that is typically about a week behind, and a feature documented this week may not have reached you yet. Homebrew’s claude-code cask tracks that stable channel too.

Write down what you decided

Reading the docs usually ends in a choice: which permission mode the team uses, which hooks are required, which MCP servers are allowed. The docs tell you the options; they cannot record which one you picked or why. On fenbs, that choice goes on the Decisions and rules page as a rule, made by a person, and every connected assistant, Claude Code included, reads the rules first through fenbs_get_context. Facts the next assistant needs, such as which settings file holds the shared config, go in AI context notes. The work that follows is a task with a note, a plan and a test status. A task-tracking workflow for Claude Code shows how that fits into a session.