Claude Code

What it is

Claude Code is a coding agent made by Anthropic. An agent here means a program that does not only answer a question but carries out a series of steps: it reads files, searches code, edits them, runs tests and can commit changes to a repository. It began as a tool that runs in the terminal, and Anthropic has since extended it to other surfaces, so that the same agent can be used from an editor, a desktop app, a web page, a phone or a chat channel.

Origins

Claude Code was announced on 24 February 2025 alongside Claude 3.7 Sonnet, as a limited research preview. In that first form it ran in the terminal and could search and edit code, run tests and commit to GitHub.^[1]

It became generally available on 22 May 2025 with Claude 4. That release added beta extensions for VS Code and JetBrains, Claude Code on GitHub Actions, and an SDK for building custom agents.^[2]

With Claude Sonnet 4.5, on 29 September 2025, it gained checkpoints, a refreshed terminal interface and a native VS Code extension. Its underlying infrastructure became the Claude Agent SDK.^[3]

Claude Code on the web launched on 20 October 2025 as a research preview for Pro and Max users. Tasks run in isolated cloud sandboxes and can end in pull requests, and the feature also appeared in the iOS app.^[4]

Growth

Anthropic reported that run-rate revenue passed $500 million by September 2025, with usage up more than tenfold in three months.^[6] Run-rate revenue is a company's current revenue multiplied out over a year. The figure reached $1 billion in November 2025, about six months after general availability.^[5]

In December 2025 Anthropic acquired Bun, a JavaScript runtime that had helped build Claude Code's native installer.^[5] By 12 February 2026 Claude Code's run-rate revenue exceeded $2.5 billion. Anthropic cited a SemiAnalysis estimate that about 4% of public GitHub commits were authored by Claude Code.^[7]

Key features

Several features sit on top of the basic agent. CLAUDE.md is a project memory file: the agent reads it to learn a project's conventions. Skills are reusable packages of instructions, hooks run commands at set points in the workflow, subagents are smaller agents that a main agent can hand a task to, and MCP servers connect the agent to outside tools and data.^[11]

The model behind the agent has also changed how the tool behaves. Agent teams arrived as a research preview with Claude Opus 4.6 in February 2026. With Opus 4.7, xhigh became the default effort level, and /ultrareview and an auto mode arrived for Max users. Opus 4.8 added dynamic workflows that can run hundreds of parallel subagents.^[8][9][10]

Today the tool runs in the terminal, in VS Code and JetBrains, as a desktop app, on the web and in mobile apps, and in Slack. Cloud scheduled tasks, which were formerly called routines, let it run work on a timetable.^[11]

Where it stands

Claude Code is one of the clearest cases of a Claude product built for software work. Its path from a terminal preview to a multi-surface tool shows how Anthropic has moved agentic features from a model's capability into a product people use daily. Wikipedia reports a leak of the Claude Code command-line source in March 2026, which is a reminder that the internals of a widely used tool can become public outside the company's control.^[12]

The tool also connects to two shared standards that appear elsewhere in this archive: MCP, the protocol for connecting models to external systems, and Agent Skills, the folder format for reusable instructions. Their presence in Claude Code is one reason the tool is often described as an open platform rather than a closed app.^[11]