Out of the box, Claude Code is a capable coding agent. With the right plugins, it becomes a disciplined engineer, a cheaper one, and one that never stops when you hit a rate limit. These five add-ons are the ones worth installing first in 2026.
Plugins arrived in Claude Code in October 2025. A plugin bundles skills, slash commands, subagents, hooks and MCP servers into a single install. You add a marketplace (a GitHub repo that lists plugins), then install from it.
# Inside a Claude Code session
/plugin marketplace add <owner>/<repo>
/plugin install <plugin-name>@<marketplace-name>
# Or from your terminal
claude plugin install <plugin-name>@<marketplace-name>Anthropic’s official marketplace, claude-plugins-official, is registered automatically the first time you start an interactive session. Type /plugin at any time to browse, enable, disable or update what you have installed.
1. OmniRoute: never stop on a rate limit
Best for: heavy daily users who keep hitting Claude Code’s usage limits mid-session.
OmniRoute is a free, MIT-licensed AI gateway with about 75,000 GitHub stars. Strictly speaking, it is not a /plugin install. It runs as a local proxy on your machine, and Claude Code sends its requests through it.
Why that matters: when your Claude subscription quota runs out, OmniRoute falls back to the next provider in your chain automatically. It supports 350+ providers, including Kimi, GLM, DeepSeek, Gemini and GPT models, many with free tiers. One independent reviewer reported that a quota switch mid-session went unnoticed until he checked the logs.
Key features:
- Combos: a chain of models it routes across, with 19 strategies such as
cost-optimized,priorityandround-robin. - Zero-config routing: set the model to
auto,auto/coding,auto/cheaporauto/subscription. - Token compression: built-in RTK and Caveman compression to shrink tool output and replies.
- Local-first: API keys stay encrypted on your machine, with a dashboard at
localhost:20128.
Install and connect:
npm install -g omniroute
omniroute # starts the gateway and dashboard
omniroute setup-claude # writes Claude Code profiles
omniroute launch # starts Claude Code through OmniRoutePrefer manual setup? Point ANTHROPIC_BASE_URL at your OmniRoute root URL (no /v1 suffix) and restart Claude Code, since it reads environment variables only at startup.
Watch out for: the free-token headline is pooled across providers you sign up for yourself, so daily headroom is smaller than it sounds. Some providers are flagged in OmniRoute’s terms-risk catalog, so read their terms. Tool calling can also be less reliable on non-Anthropic models, so test file edits before trusting a fallback model with real work.
2. Superpowers: turn Claude into a disciplined senior engineer
Best for: anyone building real features who is tired of Claude jumping straight into code.
Superpowers by Jesse Vincent is the most popular plugin in the Claude Code ecosystem, with about 293,000 GitHub stars. It launched on October 9, 2025, alongside Anthropic’s plugin system, and is MIT-licensed.
It is a full development methodology packaged as skills that trigger automatically. Instead of writing code the moment you ask, Claude first asks what you are really trying to build. The workflow then runs in order:
- Brainstorming: refines your rough idea through questions and saves a design document.
- Git worktrees: creates an isolated branch and checks for a clean test baseline.
- Writing plans: splits the work into tasks of 2 to 5 minutes each, with exact file paths and verification steps.
- Subagent-driven development: sends a fresh subagent to each task, reviewing after every one.
- Test-driven development: enforces red-green-refactor and deletes code written before its test.
- Code review: checks work against the plan; critical issues block progress.
- Finishing the branch: verifies tests, then offers merge, PR, keep or discard.
The payoff is autonomy. It is common for Claude to work for a couple of hours without drifting from the plan you approved. A systematic-debugging skill also replaces guess-and-check fixes with a four-phase root-cause process.
Install:
/plugin install superpowers@claude-plugins-officialWatch out for: the process adds time and tokens to small tasks. For a one-line fix, the brainstorm-plan-review loop can feel heavy, so it shines most on multi-file features.
3. Caveman: same answers, far fewer tokens
Best for: developers on usage-capped plans or paying per token who want shorter, faster replies.
Caveman by Julius Brussee makes Claude drop the filler and reply in blunt, telegraphic sentences. It went viral in April 2026 and now has about 103,000 GitHub stars. Code, commands, file paths and error messages are never shortened; only the prose around them is.
A typical example: a 69-token explanation of a React re-render bug becomes a 19-token reply with the same diagnosis and the same useMemo fix.
Across 10 coding prompts run against the Claude API, output tokens fell by 65% on average:
| Task | Normal (tokens) | Caveman (tokens) | Saved |
|---|---|---|---|
| Implement React error boundary | 3,454 | 456 | 87% |
| Set up PostgreSQL connection pool | 2,347 | 380 | 84% |
| Explain git rebase vs merge | 702 | 292 | 58% |
| Refactor callback to async/await | 387 | 301 | 22% |
| Average of all 10 prompts | 1,214 | 294 | 65% |
Useful commands it adds:
/caveman lite|full|ultrasets the intensity;/caveman offturns it off./caveman-commitwrites terse Conventional Commit messages./caveman-reviewgives one-line, actionable review findings./caveman-compress CLAUDE.mdshrinks memory files (about 46% fewer input tokens) and keeps a readable backup./caveman-statsshows tokens saved in your Claude Code session.
Install:
claude plugin marketplace add JuliusBrussee/caveman
claude plugin install caveman@cavemanWatch out for: the skill only shortens output. Input and reasoning tokens stay the same, and its rules add about 1,000 to 1,500 input tokens per turn. Its own maintainer says whole-session savings land lower than the table, so measure your own usage. Also, terse replies skip the “why”; switch it off when you are learning a new codebase. The optional proxy and CLI send anonymous usage stats by default (caveman telemetry off disables them).
4. Context7: stop Claude from using outdated docs
Best for: anyone working with fast-moving frameworks such as Next.js, Laravel, React or Tailwind.
Claude’s knowledge of a library is frozen at its training date. That is how you end up with deprecated APIs and config options that no longer exist. Context7, by Upstash, fixes this by fetching current, version-specific documentation through an MCP server. It has about 418,000 installs from Anthropic’s official directory.
Using it is simple. Add “use context7” to your prompt:
How do I set up authentication in Next.js 15? use context7Claude looks up the library ID, pulls the docs for that version, and answers from them instead of from memory.
Install:
/plugin install context7@claude-plugins-officialWatch out for: your queries go to Context7’s hosted server at mcp.context7.com, so avoid pasting secrets into doc questions. If you hit rate limits, set a free CONTEXT7_API_KEY.
5. Frontend Design: UIs that don’t look AI-generated
Best for: developers who build landing pages, dashboards or app screens and are tired of the same generic look.
Frontend Design is Anthropic’s own plugin and the most-installed Claude Code plugin of all, with about 1.13 million installs. Left alone, AI-built interfaces tend to converge on the same safe fonts, purple gradients and card grids. This plugin adds one skill that makes Claude commit to a clear design direction (typography, colour, layout and motion) before it writes any UI code.
It triggers automatically whenever you ask for frontend work, so there are no commands to learn:
Build a pricing page for my EV charging app with three plans.It is also cheap. It adds only about 78 tokens to each session, among the lightest of the popular plugins.
Install:
/plugin install frontend-design@claude-plugins-officialWatch out for: bold design choices are the point, so give it your brand colours, fonts or an existing design system if you need it to stay on-brand.
Quick comparison
| Plugin | What it fixes | Type | Popularity | Install from |
|---|---|---|---|---|
| OmniRoute | Rate limits and cost | Local AI gateway (npm) | ~75K GitHub stars | npm install -g omniroute |
| Superpowers | Code-first, plan-never habits | Skills plugin | ~293K GitHub stars | claude-plugins-official |
| Caveman | Wordy replies, token spend | Skills plugin | ~103K GitHub stars | JuliusBrussee/caveman |
| Context7 | Outdated library knowledge | MCP plugin | ~418K installs | claude-plugins-official |
| Frontend Design | Generic-looking UI | Skill plugin | ~1.13M installs | claude-plugins-official |
Which should you install first?
Start with Superpowers and Context7: one fixes how Claude works, the other fixes what it knows. Add Frontend Design if you ship UI. Reach for Caveman and OmniRoute once cost and rate limits become the bottleneck; they stack well, and OmniRoute even bundles Caveman-style compression.
Two tips before you go. Install each plugin from one source only; installing Superpowers from both the official and the author’s marketplace gives you duplicate skills. And every plugin adds context to each session, so run /plugin now and then to disable what you no longer use.
Which Claude Code plugins are in your setup? Tell us in the comments.
