What is Nuclexa?

Nuclexa is the package registry for AI agents and installable AI capabilities. If npm is where you get JavaScript packages and the VS Code Marketplace is where you get editor extensions, Nuclexa is where you get the installable building blocks that make AI assistants useful inside real projects — agents, skills, MCP servers, rules, workflows, prompt packs, and more.

Today these capabilities live scattered across gists, blog posts, private Notion pages, and copy-pasted system prompts. There is no shared format, no versioning, no permission model, and no way to know whether something you found online is safe to run against your codebase. Nuclexa fixes that with a standard package format, a registry to publish and discover packages, and a CLI to install them into any AI tool.

What you can do

• Discover trusted packages across every major platform — Claude Code, Claude Desktop, Cursor, Windsurf, OpenAI Agents, Gemini CLI, GitHub Copilot, Replit Agent, and any MCP-compatible client.
• Install any package with a single command into the tool of your choice — or export it to a tool's native format.
• Publish your own capabilities with semver versioning, changelogs, and a declared permission scope.
• Audit exactly what a package can do before you run it: every package surfaces its permissions, risk level, license, and source.

Core principles

1. Trust by default. Every package declares the permissions it needs. We compute a risk level from that declaration and surface it everywhere — search results, package pages, and the install dialog.
2. Portable, not platform-locked. A package is a directory of plain files, not a vendor blob. The same package installs to Claude Code, Cursor, or Gemini, and exports to Copilot or OpenAI Agents.
3. Versioned and reproducible. Packages use semver. Pin a version, read its changelog, and roll forward or back deterministically.
4. Open format. The package format is documented, simple, and yours. There is no proprietary runtime you are forced to adopt.

A package in one minute

A package is just a directory with a manifest and some content:

security-review/
  agent.json         # manifest: name, version, type, platforms, permissions
  README.md          # human-facing overview
  instructions.md    # the actual entry point loaded by the tool
  permissions.json   # declared permission scope (optional override)
  examples/          # example prompts and transcripts

Install it from the registry with the CLI:

npx nuclexa install security-review --target claude-code

New here? Read Installable AI capabilities next, then the agent.json manifest reference.