CLI + MCP server / 100% local
Every coding assistant wants its own project file in its own format. claude-init scans your repo once and writes all of them - no API key, no network.
Stack, scripts, structure, env vars, conventions and git - detected once, written to all 13 formats in seconds.
No API key, no network, nothing leaves your disk. The optional --enrich flag is the only exception, and it's off by default.
claude-init check diffs the generated files against disk and exits non-zero on drift. Wire it into CI or a pre-commit hook.
Run it over the Model Context Protocol and an agent can analyze a repo and write its own context files.
// command line
# generate everything in the current repo
npx @horiastanxd/claude-init
# only specific tools
npx @horiastanxd/claude-init -t claude,cursor
# verify files are up to date (CI / pre-commit)
npx @horiastanxd/claude-init check
# monorepo: generate into each workspace package too
npx @horiastanxd/claude-init --recurse
// github action
- uses: horiastanxd/claude-init@v1
with:
command: check
targets: all
Language, framework, runtime, package manager, database, test runner, build / monorepo tool, CI provider.
JS/TS (npm, pnpm, yarn, bun), Deno, Python (pip, uv, poetry), Rust, Go, Java/Kotlin, C#/.NET, Elixir, Dart, Ruby, PHP.
Next.js, Remix, Astro, Angular, SvelteKit, Nuxt, Vue, React, Fresh, FastAPI, Django, Rails, Laravel, Spring Boot, ASP.NET Core, Phoenix, Flutter, and more.
Strict mode, linter, formatter, import style, commit convention, env vars, git authors and hot files.
Q.What is claude-init?
One command that scans a repository and generates AI context files - CLAUDE.md, AGENTS.md, Cursor, Windsurf, Cline, Continue, Kilo Code, Trae, Copilot, GEMINI.md, Aider, Junie and Warp - for every major coding assistant. It runs as a CLI or an MCP server, 100% local.
Q.How do I generate a CLAUDE.md file?
Run npx @horiastanxd/claude-init in your project. It detects your stack, scripts, structure and conventions, then writes CLAUDE.md alongside every other tool's context file. No config and no API key required.
Q.What is the difference between CLAUDE.md and AGENTS.md?
CLAUDE.md is read by Claude Code. AGENTS.md is an emerging shared standard read by Codex, Jules, Amp, Zed, Devin and 20+ other agents. claude-init writes both, so your project is covered whichever tool you use. Read the full comparison →
Q.Does claude-init send my code anywhere?
No. It runs entirely on your machine - no API key and no network. The only exception is the optional --enrich flag, which uses an LLM to sharpen the description and is off by default.
Q.How do I keep AI context files up to date?
Run claude-init check in CI or a pre-commit hook. It regenerates the files in memory, diffs them against what is on disk, and exits non-zero on drift so stale context fails the build.
Q.Which languages and frameworks does claude-init detect?
JavaScript/TypeScript, Deno, Python, Rust, Go, Java/Kotlin, C#/.NET, Elixir, Dart, Ruby and PHP - with frameworks including Next.js, Astro, FastAPI, Django, Rails, Laravel, Spring Boot, ASP.NET Core, Phoenix and Flutter.
Q.Is claude-init free?
Yes. claude-init is free and open source under the MIT license.