SuperClaude is a configuration framework that layers behavioral instructions and orchestrated components onto Claude Code. Version 4.x installs a set of /sc: slash commands spanning the development lifecycle, from brainstorming and research through implementation, testing, and project management, together with specialized agents, behavioral modes, and optional MCP server integrations. It is not affiliated with Anthropic. A TypeScript plugin system is planned for v5.0 but not yet available.
The framework ships 30 slash commands, 20 agents (for example a PM agent, a deep research agent, a security engineer, and a frontend architect) that coordinate automatically based on context, and 7 behavioral modes. Eight MCP servers can be installed through the superclaude mcp command; the project reports that Serena and Sequential reduce execution time and token use, while the framework remains functional without any of them.
The Deep Research capability runs autonomous web research with three planning strategies (planning-only, intent-planning, and unified), up to five iterative search hops, confidence-based quality scoring with a 0.6 minimum and 0.8 target, and case-based learning across sessions. Four research depth levels trade sources and hops against time.
| Depth | Sources | Hops | Time |
|---|---|---|---|
| Quick | 5–10 | 1 | ~2 min |
| Standard | 10–20 | 3 | ~5 min |
| Deep | 20–40 | 4 | ~8 min |
| Exhaustive | 40+ | 5 | ~10 min |
Features
- Slash commands: 30 /sc: commands including research, brainstorm, implement, test, and pm
- Specialized agents: 20 domain agents that coordinate based on context
- Behavioral modes: brainstorming, business panel, deep research, orchestration, token-efficiency, task management, and introspection
- MCP servers: Tavily, Context7, Sequential-Thinking, Serena, Playwright, Magic, Morphllm-Fast-Apply, and Chrome DevTools, installed via the CLI
- Deep research: adaptive planning, multi-hop reasoning, source credibility scoring, and cross-session learning
- Project documents: PLANNING.md, TASK.md, and KNOWLEDGE.md that Claude Code reads at session start
- Diagnostics: superclaude doctor and superclaude install --list verify the setup
- Sibling frameworks: SuperGemini and SuperQwen apply the same approach to other coding agents
Integrates with
Context7
Fetches up-to-date, version-specific library docs and code examples into a coding agent's context via CLI or MCP
Playwright MCP
MCP server giving LLMs Playwright browser automation through structured accessibility snapshots rather than screenshots
Serena
On a large codebase, a coding agent reads whole files and edits by text match
Tavily MCP
MCP server exposing Tavily search, extract, map, and crawl tools, run locally or via a hosted remote endpoint