5 Claude Code Plugins Worth Installing in 2026 (And the Fine Print on the Unlimited Tokens Claim)

Laptop displaying programming code on wooden desk with coffee cup and plant at night

A YouTube breakdown from Nick Automates promises five Claude Code plugins that hand you unlimited tokens and a 10x workflow. Four of the five hold up once you check the source code and the docs. The fifth comes with a real security history and a token number that needs context before you trust it.

Claude Code plugins have exploded since Anthropic opened up the system in late 2025, and the pitch in most “top plugins” videos is the same: install five things, stop hitting limits, code faster. Some of that pitch is fair. Some of it is marketing math dressed up as a fact. Here’s what each of these five plugins actually does, how to install it, and where the video’s framing needs a correction.

The Five at a Glance

PluginWhat it solvesInstall effortWatch out for
Claude Code SetupRecommends hooks, skills, MCP servers, and subagents for your codebaseOne commandNothing, it’s official and read only
Claude MemRemembers project context across sessionsTwo commandsStores session data locally, review what it captures
HeadroomCompresses tool output before it reaches the modelOne pip installReal savings are lower than the headline number
Task ObserverLogs corrections and turns them into skill upgradesCopy a skill folderIt’s a skill, not a marketplace plugin, and works best in Claude Cowork
Omni RouteRoutes Claude Code to free third-party models when you hit a limitOne npm install, several security stepsDefault config is insecure, and the token math is a ceiling, not a guarantee

1. Claude Code Setup: Let Anthropic Map Your Automations First

This is the one plugin in the list that is actually built and maintained by Anthropic, not a community project. It lives in the official claude-plugins-official marketplace, and its job is narrow: it scans your codebase and tells you which hooks, skills, MCP servers, and subagents would actually help, instead of you guessing from a list of 70-plus community plugins.

It reads your project. It does not touch it. That distinction matters once you start installing plugins that do modify files.

How to install it:

  1. Open a terminal in your project and start Claude Code.
  2. Run /plugin marketplace add anthropics/claude-plugins-official.
  3. Run /plugin install claude-code-setup@claude-plugins-official.
  4. Ask Claude something like “analyze this codebase and recommend automations.” The skill scans your project and returns its top one or two picks in each category: MCP servers for docs or browser testing, skills for design or planning, hooks for auto-formatting or blocking sensitive files, subagents for security or performance review, and slash commands for common workflows.

Install this one first. It will point you toward plugins that fit your specific project instead of whatever a YouTube video happened to feature that week.

2. Claude Mem: Stop Re-Explaining Your Project Every Session

Every new Claude Code session starts from zero. You explain your stack, your conventions, the bug you fixed last Tuesday, again. Claude Mem, built by a developer known as thedotmack, fixes that by capturing what happens during a session, compressing it with Claude’s own agent SDK, and storing it in a local SQLite database that future sessions can pull from automatically.

How to install it:

  1. In Claude Code, run /plugin marketplace add thedotmack/claude-mem.
  2. Run /plugin install claude-mem@thedotmack.
  3. Restart Claude Code.
  4. Work as usual. On your next session, relevant context from past work should surface automatically, no re-explaining required.

A few features worth knowing about: mem:search lets you query past sessions by keyword, mem:forget removes specific memories you don’t want kept, and wrapping sensitive text in <private> tags keeps it out of storage entirely. There’s also a local web viewer if you want to see exactly what it’s captured. Since this stores a running record of your work, it’s worth a quick look at what’s landing in there if you work with client code or anything under an NDA.

3. Headroom: Cut Token Waste From Bloated Tool Output

A lot of what burns through your Claude Code budget isn’t the model thinking, it’s the model reading. A 100-line JSON API response, a sprawling build log, a directory listing you didn’t need in full: all of that gets read in full unless something trims it first. Headroom, built by Tejas Chopra, sits between your agent and the model and rewrites that noisy content down to what actually matters, while keeping the ability to pull the original back if the model needs it.

How to install it:

  1. Run pip install "headroom-ai[all]" (Python 3.10 or later; Node and Docker installs are also available).
  2. Run headroom wrap claude to launch a Claude Code session through Headroom’s compression layer.
  3. Keep that terminal window open for the session. Compression happens automatically in the background.
  4. Run headroom perf or headroom dashboard to see what it actually saved on your own work, not the marketing number.

That last step matters. Headroom’s own benchmarks show 60 to 95 percent token reduction on log- and JSON-heavy content, which is real and reproducible. But that’s the best case, not the average session. On a normal coding workload, where plenty of turns have little worth compressing, published third-party testing puts realistic savings closer to 15 to 25 percent overall. Still worth having. Just don’t expect your weekly usage bar to move by 90 percent.

4. Task Observer: Turn Corrections Into Skill Upgrades Automatically

One correction worth flagging up front: this is a skill, not a marketplace plugin in the formal sense. It’s part of a project called one-skill-to-rule-them-all, built by Eoghan Henn, and it does something genuinely useful: it watches your work sessions and logs the moments where you correct Claude, repeat a workflow, or find a technique that worked well. Those observations feed a review cycle that turns patterns into new or improved skills, including improving itself.

How to install it:

  1. Grab the skill folder from the GitHub repository.
  2. In Claude Code, place it at .claude/skills/task-observer/ for a single project, or in your user-level skills directory to use it everywhere. Keep the references/ subfolder intact, it’s part of the skill.
  3. Work normally. It starts logging to skill-observations/log.md the first time it’s triggered.
  4. At the end of a session, ask “any observations logged?” to see what it caught, or wait for a scheduled review to turn the log into concrete skill updates.

The developer behind it reports around 600 skill improvements applied across 40 personal skills over three months of daily use, which gives you a sense of how much a session can quietly generate once you’re not the one tracking it. It’s built to work fully in Claude Cowork and the Claude Code desktop app, where it writes persistent logs to your filesystem. In Claude.ai on web or mobile, it falls back to a lighter handoff-note mode instead.

5. Omni Route: The Free-Model Router (Read This Before You Install It)

This is the plugin doing the heavy lifting behind that “unlimited tokens” headline, and it deserves a slower look than the other four.

Omni Route is a local gateway that sits between Claude Code and the internet. It catalogs 290-plus AI providers, more than 90 of which offer some kind of free tier, and when your primary model hits a rate limit, it automatically falls back to the next available option. It also runs your tool output through a compression pipeline the project calls RTK and Caveman, which it claims saves 15 to 95 percent of tokens depending on content.

If you’ve read our guide to running Claude Code through a free-model proxy, this will sound familiar. The core trick is the same: your Claude Code interface stays put, and the model underneath gets swapped. Omni Route’s version just catalogs a much wider net of providers, and bundles in the token compression that Headroom does on its own.

How to install it:

  1. Run npm install -g omniroute (Docker and a desktop Electron app are also available).
  2. Before starting the server, set your own JWT_SECRET, API_KEY_SECRET, and STORAGE_ENCRYPTION_KEY environment variables. Do not skip this step, more on why below.
  3. Start it. It boots a local server at http://localhost:20128.
  4. Point Claude Code at http://localhost:20128/v1. A handful of keyless free providers are pre-wired, so a fresh install responds immediately with no setup, and you can add your own free-tier API keys through the local dashboard as you go.

Now, the honest caveats.

The “1.6 billion free tokens” figure comes from the project’s own dashboard, which currently states roughly 1.53 billion. It is not a guarantee, and the project’s own documentation says so directly: it’s the sum of every provider’s documented rate limit extrapolated out to nonstop use across an entire month. You will not personally pull down anything close to that number. It’s a theoretical ceiling across the whole catalog, not a balance sitting in your account.

The security history is worth knowing too. Omni Route’s predecessor project shipped a hardcoded default authentication secret, which led to a critical vulnerability that let attackers bypass login entirely and pull stored API keys. Omni Route’s own security documentation acknowledges the same risk exists if you don’t set your own secrets, which is exactly why step two above isn’t optional. Separately, in May 2026, the security scanner Socket.dev flagged the npm package over suspicious install-script behavior. The maintainer responded by patching two confirmed vulnerabilities in the next release, and independent reviewers rated that response as responsible. No malware was ultimately confirmed. Still, this is a single-maintainer, community-run project handling your API keys on your machine, not an Anthropic product.

None of that means skip it. It means treat it like what it is: a genuinely useful tool for stretching free-tier access, run with a fresh set of API keys you wouldn’t miss, never pointed at anything tied to production, billing, or an account you can’t afford to lose.

How These Five Actually Fit Together

You don’t need all five running at once, and installing every plugin you find is how you end up with a Claude Code session slower than the one you were trying to fix. A reasonable starting stack looks like this: install Claude Code Setup first and let it tell you what your project actually needs, add Claude Mem if you’re tired of re-explaining context every morning, add Headroom or Omni Route’s compression if your token usage runs hot, and add Task Observer once you notice you’re correcting Claude the same way more than once. Omni Route earns its place specifically when you’re hitting hard limits on a budget, not as a default.

Unlimited tokens is a headline. What these plugins actually give you is fewer wasted ones, and that’s worth more than the marketing math.

Sources and Further Reading

  1. Claude Code Setup, official Anthropic plugin marketplace
  2. Claude Mem on GitHub
  3. Headroom on GitHub
  4. Headroom on PyPI (install and benchmark details)
  5. Task Observer / one-skill-to-rule-them-all on GitHub
  6. Omni Route on GitHub
  7. Omni Route’s own free-tier token methodology
  8. Independent security review of Omni Route’s token claims and CVE history
  9. Original video: Top 5 Claude Code Plugins, Nick Automates

Leave a comment

Website Built by WordPress.com.

Up ↑