Anthropic official marketplace plugin
What it does
Claude Code exposes five extensibility surfaces — hooks, skills, MCP servers, subagents, and slash commands — each with dozens of viable options. Picking the right combination from cold is hard.
This plugin scans the codebase read-only and surfaces the top 1–2 automations per category, grounded in real signals from package.json, framework choices, database, CI, and issue tracker references — rather than the generic shortlist a blog post would suggest.
How it works
Phase 1 — codebase analysis. The skill inspects package.json, pyproject.toml, Cargo.toml, go.mod, framework imports, dependency lists, and existing .claude/ / CLAUDE.md configuration. It captures language and framework, frontend and backend stack, database, external APIs (Stripe, OpenAI, AWS SDKs, etc.), testing setup, CI/CD, and issue tracker references.
Phase 2 — recommendations. Based on what it found, it generates ranked recommendations across all five categories. By default it returns the top 1–2 per category to avoid overwhelming the output. Ask for a specific category to widen it to 3–5.
What it recommends
| Category | Example recommendations |
|---|---|
| MCP servers | context7 (live docs), Playwright (browser/UI testing), Supabase / PostgreSQL MCP, GitHub MCP, Linear MCP |
| Skills | frontend-design, commit, feature-dev, plus custom skill templates scoped to the codebase |
| Hooks | PostToolUse auto-format, lint on edit, block sensitive files, run tests pre-commit |
| Subagents | Specialized reviewers — security, performance, accessibility |
| Slash commands | /test, /pr-review, /explain and similar quick workflows |
Each recommendation comes with a short rationale tying it back to a specific signal in the codebase, so you can adopt the full list or cherry-pick.
Use cases
- First time on a fresh project — Run
recommend automations for this projectto get a full five-category diagnostic in one pass. - Going deeper on one category — Ask “recommend more MCP servers for this codebase” to expand a single category to 3–5 candidates.
- Switching codebases — Re-run the diagnostic on a new repo instead of copying your previous setup wholesale, so unused automations don’t tag along.