Anthropic's Claude Code hits 119K GitHub stars: why developers are skipping the IDE

If you work primarily in the terminal, Claude Code is probably the most relevant AI coding tool I've seen built for the way you already work. Anthropic's terminal-based coding agent just crossed 119K GitHub stars and 19.8K forks, and unlike a lot of starred repos, the development activity backs it up: 607 commits, 52 contributors, and a changelog updated as recently as April 28, 2026.

The repository has come a long way from its research preview. It hit general availability with version 1.0.0 in May 2025, and what's been built since is worth a closer look if you haven't revisited it lately.

What Happened

Claude Code reached general availability in May 2025 and has been in steady development since. The tool accepts natural-language commands and operates across the terminal, IDEs, and GitHub (via @claude mentions on PRs and issues).

One thing I'd flag here: Anthropic has moved away from npm installation entirely. The older npm install -g @anthropic-ai/claude-code method is now deprecated. They now recommend native installers for macOS, Linux, and Windows, including support for Homebrew and WinGet. If you tried it early and hit friction during setup, that experience is likely different now.

The pace of commits and the April 2026 changelog updates tell me this isn't a tool Anthropic shipped and moved on from. Active plugin development, MDM deployment templates, and devcontainer support are all landing in the repo. It's being invested in, which matters when you're evaluating whether to build a team workflow around something.

Key Features

• Terminal-native agent. Runs in your shell with a single claude command from your project directory. No browser tab, no context switching.
• Natural language git workflows. Handles commits, diffs, and PRs through plain English. You can also tag @claude directly on GitHub to trigger actions on PRs and issues without opening a terminal session.
• Cross-platform install. Available via curl on macOS and Linux, irm on Windows, Homebrew, and WinGet. Setup is straightforward in a way that early versions weren't.
• Plugin system. The repo includes a plugins directory that ships with a code review plugin that posts inline PR comments and a learning-output-style plugin. Custom commands and agents extend the core functionality: this is the part I'd spend the most time exploring.
• IDE integration. Works in VS Code via a dedicated extension, with devcontainer support for remote development environments.
• MCP tool support. The code review plugin uses Model Context Protocol tools with confirmation flags to prevent subagent comments from leaking into customer PRs.

Why It Matters

Terminal-based AI tools solve a real friction problem. Instead of copying code into a chat window, switching contexts to a browser, and copying the output back, you stay in your shell, where your repo, your git history, and your actual context already live. Claude Code sits in that workflow and acts on it directly.

The plugin architecture is where I think the longer-term story is. Anthropic is clearly building an extensible platform here, not just a chatbot wrapper. The code review plugin posting inline PR comments is a concrete example of what that looks like in practice: a workflow that slots into existing processes rather than asking teams to adopt a new one.

I'm seeing AI coding tools move away from autocomplete and toward agents that actually execute work. Claude Code is one of the clearer examples: you describe a task, and it acts on your codebase. For senior engineers who are already shell-centric, the adoption ask is genuinely low. There's no new UI to learn, no editor to switch to. You type claude and start prompting against your actual codebase.

Example Use Case

A backend team maintaining a Python and TypeScript monorepo needs to update error handling across 40+ API endpoints. Instead of manually editing each file, a developer runs claude in the project root and describes the task in plain English: add structured error responses with status codes to all route handlers that currently raise raw exceptions.

Claude Code reads the codebase, identifies the relevant files, applies the changes, and stages the git diff. The developer reviews the output in the terminal, adjusts two edge cases, and commits. The plugin system could further automate this by adding a custom command that enforces the team's error-handling conventions on every PR.

This is the workflow I'd demo to a skeptic. Code comprehension, file manipulation, and git operations from a single terminal session. That combination is harder to replicate with a standard IDE extension.

Competitive Context

Claude Code is Anthropic's answer to a specific question: what does Claude look like when it has direct access to your codebase and git? The conversational model handles general-purpose tasks through chat interfaces. Claude Code wires that same intelligence into your terminal. The difference is between asking a model to write code in a chat window and having an agent that executes changes in your actual repo.

What I find more interesting than the terminal positioning is the open plugin system. Proprietary IDE tools keep their agent infrastructure closed. Claude Code is publicly available, inspectable, and contributable. That's a meaningful difference for teams that want to extend or audit what their AI tooling is actually doing.

The @claude tagging on GitHub adds a collaboration layer directly in the repository, which is a low-friction way to get AI assistance without requiring every developer to change their local setup. At 19.8K forks and 52 contributors, the project has healthy community involvement alongside Anthropic's own development effort.

My Take

If your team works in the terminal and wants AI assistance without leaving it, Claude Code is worth a serious evaluation. The 119K stars reflect a tool that fits how many developers already work, and the plugin architecture means it's not limited to what Anthropic ships by default.