Claude Code

Claude Code

by Anthropic

Agentic coding tool that reads your codebase, edits files, runs commands and integrates with your dev tools, from terminal, IDE, desktop app and browser.

(Filename kept from the 2025 “Claude Code 2.0” note; this is the general Claude Code note.)

Summary

Claude Code is Anthropic’s coding agent. One engine powers every surface, so CLAUDE.md files, settings and MCP servers carry across them. Claude Code 2.0 (autumn 2025) introduced the VS Code extension, checkpoints and the rebranded Agent SDK; it has since grown into a full platform with plugins, skills, background agents, scheduled routines and cloud sessions.

Surfaces

• Terminal CLI: the full-featured surface; native installer, Homebrew, WinGet, apt/dnf/apk.
• IDE extensions: VS Code (also installable in Cursor) and a JetBrains plugin (requires the CLI).
• Desktop app: standalone app (macOS, Windows; Linux beta) with visual diff review, parallel sessions, scheduled tasks. Paid subscription required.
• Web and mobile: claude.ai/code and the Claude iOS/Android apps; claude --teleport pulls a web session into the terminal.
• Also: Slack (@Claude), GitHub Actions / GitLab CI/CD, Chrome debugging, Remote Control from phone.

Features

• Memory: CLAUDE.md (managed, user, project, local scopes; .claude/rules/ with path-scoped rules) plus auto memory that Claude writes itself. AGENTS.md: read directly when no CLAUDE.md/CLAUDE.local.md exists in or above the working directory (v2.1.277+), otherwise import with @AGENTS.md inside CLAUDE.md. See agents-md.
• Skills: SKILL.md folders following the open Agent Skills standard; custom slash commands have been merged into skills (old .claude/commands/ files still work).
• Hooks: deterministic shell/HTTP/MCP-tool/prompt/agent handlers on lifecycle events (PreToolUse, PostToolUse, Stop, SessionStart and many more); exit code 2 blocks.
• Plugins: a directory bundling skills, subagents, hooks and MCP servers, installed from marketplaces via /plugin.
• Subagents and parallel/background agents; checkpoints and /rewind.
• MCP servers for external tools and data.
• Automation: headless claude -p, /loop, cloud Routines (/schedule), GitHub/GitLab CI.
• Agent SDK (renamed from Claude Code SDK): Python claude-agent-sdk and TypeScript @anthropic-ai/claude-agent-sdk, bundling the Claude Code binary. Third parties may not offer claude.ai login for SDK-based products.

Models

Per the changelog, v2.1.284 added Claude Sonnet 5.5 (claude-sonnet-5-5; 1M context) as the default Sonnet model. Secondary trackers report Opus 5.5 support from v2.1.280; the changelog summary did not confirm this. Also works via Bedrock, Vertex, Foundry.

Plans

Included in paid Claude plans: Pro, Max (5x or 20x usage levels), Team and Enterprise; usage draws from shared plan limits. Also usable with an Anthropic Console API key (pay per token). Current prices: see https://claude.com/pricing.

Version and cadence

Latest in the GitHub changelog on 2026-09-30: 2.1.286 (2.1.285 and 2.1.284 just before). Several releases per week; the Homebrew claude-code cask tracks a stable channel about a week behind @latest.

Known limitations and notes

• CLAUDE.md is context, not enforcement; use hooks or permission settings for hard rules.
• Plugins run code as you; review before installing.
• Checkpoints complement, not replace, version control.

History and 2025 snapshot (merged 2026-10-02 from the older claude-code note)

Attributed to the owner’s earlier Claude Code note (last modified 2025-12-30); not re-verified on 2026-10-02.

• Timeline: first internal prototype 2024-09; dogfooding version 2024-11; public research preview announced 2025-02-24 alongside Claude 3.7 Sonnet (Claude 3.7 Sonnet); Claude Code 2.0 released 2025-09-29 with Claude Sonnet 4.5 (checkpoints, subagents, hooks); Dec 2025 added background agents, named sessions (/rename, /resume), the .claude/rules/ directory and Claude in Chrome (beta). Earlier 2025 updates it lists: customizable status line, custom subagents and /export (July-August), tool renaming and queued messages (April).
• Checkpoints: code state is saved before each change; rewind with Esc twice or /rewind (code, conversation or both).
• MCP: acts as MCP client and can run as an MCP server; output warning threshold of 10,000 tokens, limit configurable via MAX_MCP_OUTPUT_TOKENS; enterprise managed-mcp.json. See MCP.
• Install (2025): npm install -g @anthropic-ai/claude-code, version 2.0.76 in Dec 2025 (the current note above lists newer install methods and versions).
• VS Code extension (2025 description): plan review before accepting, auto-accept edits, @-mention files with line ranges, inline diffs, multiple tabs.
• 2025 comparisons (author’s view, vendor-reported): terminal-native versus IDE-centred tools such as Cursor; a 77.2% SWE-bench Verified score was reported for Claude Sonnet 4.5 (vendor figure from the 2025 release, not re-checked).
• 2025 plans (superseded by the Plans section above): Free tier had no Claude Code access; Max came in two usage levels; Team had premium seats.

Related

claude-agent-sdk, claude-code-extensibility-guide, claude-code-agents-convention, claude-desktop

Sources

• https://code.claude.com/docs/en/overview (accessed 2026-09-30)
• https://code.claude.com/docs/en/memory (2026-09-30)
• https://code.claude.com/docs/en/skills, /hooks, /plugins, /agent-sdk/overview, /agent-sdk/quickstart (2026-09-30)
• https://raw.githubusercontent.com/anthropics/claude-code/main/CHANGELOG.md (2026-09-30)
• https://claude.com/pricing (2026-09-30)
• https://releasebot.io/updates/anthropic/claude-code (secondary, Opus 5.5 claim; 2026-09-30)