Claude Code Ultimate Guide
Start here: complete a first task online · browse the guide portal · open the complete sitemap · read the Markdown source
Choose your next step
The table below is generated from machine-readable/navigation.json. The same contract feeds the public sitemap, so the repository and website expose the same intent model.
Why this guide exists
Claude Code documentation explains the product. This guide connects product behavior to engineering decisions: what belongs in context, when to use an agent instead of a skill, how to verify generated work, and which controls matter when usage moves beyond one developer.
The guide favors explicit trade-offs and verifiable procedures. Where evidence is incomplete, the relevant page should preserve that limit instead of presenting one workflow as universal.
Start
Install Claude Code
Choose one installation method:
# npm, macOS, Linux, or Windows
npm install -g @anthropic-ai/claude-code
# macOS with Homebrew
brew install claude-code
# macOS or Linux native installer
curl -fsSL https://claude.ai/install.sh | sh
Windows PowerShell also supports:
irm https://claude.ai/install.ps1 | iex
Verify the installation and authenticate:
claude --version
claude doctor
claude auth login
The Quick Start chapter documents installation alternatives, authentication, updates, permission modes, and common first-day failures.
Complete a first task
Open a small repository with a clean or understood Git state, then start Claude Code:
cd your-project
claude
Give Claude a bounded request that includes the expected result and verification:
Explain how this project runs its tests. Do not modify files.
Name the relevant commands and cite the files that define them.
Before asking Claude to edit code, add a project-level CLAUDE.md that records the commands and constraints Claude must follow:
# Project instructions
## Commands
- Test: `npm test`
- Lint: `npm run lint`
## Boundaries
- Do not edit generated files.
- Do not change dependencies without approval.
- Run the relevant tests before claiming completion.
Continue with the first workflow or use the starter CLAUDE.md templates.
Choose a learning path
For a personalized route, use the repository onboarding prompt:
claude "Fetch and follow the onboarding instructions from: https://raw.githubusercontent.com/FlorianBruniaux/claude-code-ultimate-guide/main/tools/onboarding-prompt.md"
Build
Agent engineering
An agent is one component of a larger system. The surrounding harness controls context, tools, permissions, state, stopping conditions, recovery, and evaluation.
Context and memory
Context quality affects every tool call and decision. Start with project instructions, add specialized context only when a recurring task needs it, and test whether the extra material changes behavior.
Workflows
Browse every workflow for task management, GitHub Actions, production reliability, event-driven agents, design-to-code, PDF generation, search, and team instructions.
Choose the smallest interface
Claude Code can call local commands, skills, agents, hooks, plugins, and MCP servers. More infrastructure adds setup, permissions, failure modes, and maintenance.
Use the MCP or CLI decision guide and the trade-off framework before introducing another integration.
Scale
Reliability and security
Security claims and threat counts change as sources are added or corrected. Use the linked database and guides as the current source instead of copying their counts into project documentation.
Evaluation and operations
A green structural check proves only what it inspected. Runtime behavior, task acceptance, security boundaries, and business outcomes require separate evidence.
Organization and economics
The economics pages separate observed costs from estimates and scenarios. Recalculate them with your workload, acceptance criteria, review effort, and risk constraints.
Use the format that fits the task
MCP Server: Use the guide from any coding client
Install the guide as a stdio MCP server and query it from Claude Code, Codex, Cursor, VS Code, or another MCP client.
claude mcp add --scope user claude-code-guide -- npx -y [email protected]
codex mcp add claude-code-guide -- npx -y [email protected]
Project-scoped Claude Code configuration belongs in .mcp.json:
{
"mcpServers": {
"claude-code-guide": {
"type": "stdio",
"command": "npx",
"args": ["-y", "[email protected]"]
}
}
}
The list operations and search index use bundled content. Section, example, cheatsheet, changelog, digest (get_digest), and threat tools can fetch GitHub content and write a 24-hour local cache. The official-doc initialization and refresh tools fetch Anthropic documentation and write separate local snapshots.
Canonical technical guide, installation, privacy, limitations, and dated statistics
Package README and diagnostics
Companion commands rendered from the repository: /ccguide:daily, /ccguide:diff-docs, /ccguide:init-docs, /ccguide:refresh-docs, /ccguide:search-docs.
Load the guide into another assistant
Use the small index for discovery:
curl -sL https://raw.githubusercontent.com/FlorianBruniaux/claude-code-ultimate-guide/main/machine-readable/llms.txt
Use reference.yaml when the assistant needs structured topic routes and links into the full guide. Maintenance instructions belong in the machine-readable documentation, not in this README.
Complementary ecosystem
No single repository needs to cover learning, installation, curation, reference material, and every specialized interface. Choose the resource that matches the current task.
This short list was reviewed on 2026-08-31. Project activity, installation methods, and scope can change. Browse the public Ecosystem and Compare pages first. The AI Ecosystem source, Third-Party Tools source, and Resource Evaluations retain the deeper evidence.
For desktop workspaces over existing agent CLIs, see the October 2026 workspace comparison, including Conductor's local/cloud workflow and Vibe Kanban's maintenance note.
Updates
Guide changes and Claude Code product releases answer different questions:
Recent guide additions include:
- Mistral Large 4 evaluation: October 6 launch evidence, promotional pricing, independent results, regional inference, and pending weights; integrated into the market snapshot and AI ecosystem.
- Loop & Graph Engineering (source): bounded feedback, workflow state, stopping rules, recovery, and judgment allocation.
- Subscription Strategy (source): seats, APIs, gateways, self-hosting scenarios, and team-scale decision gates.
- Cross-Session Messaging (source): peer discovery, delivery, security boundaries, and correlated-drift controls.
Use the changelog for the complete history. The README should expose only a small current selection.
Languages and translations
English is the canonical edition. French is maintained in this repository. Simplified Chinese, Ukrainian, and Latin American Spanish are independent community editions with separate update schedules.
The translation status page records maintainers, source commits, coverage, measured lag, and review status.
Repository map
guide/
├── ultimate-guide.md Complete reference
├── learning-path/ Guided course
├── core/ Architecture, tools, context, methods
├── security/ Hardening, isolation, privacy, governance
├── ecosystem/ Agents, MCP, tools, inference, knowledge
├── roles/ Adoption, learning, evaluation, careers
├── ops/ Observability, metrics, economics, gateways
├── workflows/ Task-oriented delivery guides
└── diagrams/ Mermaid diagrams with text fallbacks
examples/ Reusable agents, skills, hooks, and workflows
machine-readable/ Structured indexes and release data
mcp-server/ Packaged guide access for MCP clients
quiz/ Local knowledge validation
docs/resource-evaluations/ Reviewed external resources
whitepapers/ French and English long-form sources
The detailed guide index, examples catalog, and tool index are the canonical navigation pages for those directories.
Contributing
Corrections, missing evidence, broken links, examples, quiz questions, and documentation improvements are welcome. Read CONTRIBUTING.md before opening a pull request.
Useful contribution routes:
- Report an incorrect or outdated claim through GitHub Issues.
- Submit an external resource through the resource evaluation process.
- Propose a focused guide or workflow with its sources, boundaries, and verification method.
- Improve an example without weakening its existing security checks.
Maintainer
Florian Bruniaux maintains this guide.
Explore the ecosystem
These projects extend the workflow without duplicating this tool:
- Install with claude-code-plugins: move from explanations to installable templates.
- Validate with ctxharness: test whether documented context and instructions still match the repository.
- Secure with agentsec-triage: connect security guidance to a versioned operational feed.
- Continue with Claude Cowork Guide: switch to workflows designed for knowledge workers and non-developers.
Browse the complete open-source galaxy
License
The guide uses CC BY-SA 4.0. Reusable templates use CC0 1.0. Check the repository license files before redistributing mixed content.
Watch GitHub Releases or subscribe to the RSS feed for updates.
Version 3.44.1 | Updated daily · Oct 17, 2026