Skip to content
← Tools / Skills
Skill · .md packageSuperClaude-Org/10 trustVerified
SuperClaude logo

SuperClaude

Configuration framework that supercharges Claude Code with slash commands, cognitive personas, and dev methodologies

Visit official site →

About

SuperClaude is a meta-programming configuration framework for Claude Code — a skill-style package that enhances Claude Code with 30 specialized slash commands (/sc:research, /sc:implement, /sc:test, /sc:pm, ...), 20 domain-specialist agents, 7 behavioral modes (brainstorming, deep research, orchestration, token-efficiency, ...), and optional integrations with 8 MCP servers (Context7, Sequential-Thinking, Serena, Playwright, Magic, Morphllm, Tavily, Chrome DevTools).

It is behavioral instruction injection into Claude Code, not a standalone runtime: it installs framework files, commands, agents, and settings into your '~/.claude/' configuration. The current stable version is v4.3.0; a v5 TypeScript plugin system is announced with no ETA. Note the requirements: you need Claude Code itself installed and authenticated (a separate paid product), and optional MCP servers require their own API keys. We did not hands-on verify the exact command/agent counts or the claimed token-savings — those come from the project's docs.

What it can do

  • • 30 specialized slash commands (/sc:research, /sc:brainstorm, /sc:implement, /sc:test, /sc:pm, ...) for structured dev workflows
  • • 20 domain-specialist agents (product, security, research, frontend, and more)
  • • 7 behavioral modes including brainstorming, deep research, orchestration, and token-efficiency
  • • Optional integration with 8 MCP servers: Context7, Sequential-Thinking, Serena, Playwright, Magic, Morphllm, Tavily, Chrome DevTools
  • • 'superclaude doctor' diagnostics and 'superclaude install --list' to audit installed components
  • • Local-only metrics in '.superclaude_metrics/' (no telemetry phone-home found)
  • • v5 TypeScript plugin system in development — announced, no ETA

Setup

  1. Install and authenticate Claude Code first (separate product, paid subscription).
  2. Install Python 3.10+ and pipx.
  3. Install from PyPI: 'pipx install superclaude'. (Alternative: 'git clone https://github.com/SuperClaude-Org/SuperClaude_Framework.git && cd SuperClaude_Framework && ./install.sh'.)
  4. Install the commands into Claude Code: 'superclaude install' (installs all 30 slash commands).
  5. Optionally install MCP servers: 'superclaude mcp --list' to see available servers, 'superclaude mcp' for the interactive wizard, or 'superclaude mcp --servers tavily --servers context7' for specific ones.
  6. Verify with 'superclaude install --list' and 'superclaude doctor', then restart Claude Code. Use '/sc' to list all commands; try '/sc:research', '/sc:brainstorm', '/sc:implement', '/sc:test', '/sc:pm'.

Prerequisites

Claude Code installed and authenticated (paid subscription); Python 3.10+ with pipx (or pip); Git for the direct-install route; optional third-party API keys for MCP servers (e.g. Tavily).

Use cases

codeautomation

Works with

Claude Code

Trust score — 8/10

/10 trustVerified

Scored 8/10 (verified tier, metrics-v1). Evidence: 23,909 GitHub stars (log-scaled vs catalog peers); blended with ~5,193 monthly package downloads; last push 15d ago; 6 releases/12mo; 71 open issues. Manual review: writes into ~/.claude/; optional MCP servers need third-party keys; community org, no vendor backing; 2-3x perf claims unmeasured. License: MIT.

★ 23,909 GitHub stars

How the trust score is built

Full methodology →
Adoption · stars/downloads, log-scaled vs peers
7.5/10
Maintenance · recency, releases, issue load
8.8/10
Openness · license + docs
10.0/10
Manual review · sanity read + red-flag check
7.0/10
  • · 23,909 GitHub stars (log-scaled vs catalog peers)
  • · blended with ~5,193 monthly package downloads
  • · last push 15d ago; 6 releases/12mo; 71 open issues
  • · manual review: writes into ~/.claude/; optional MCP servers need third-party keys; community org, no vendor backing; 2-3x perf claims unmeasured

Sources

Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.