10 Claude Code Plugins That Actually Matter in 2025
10 Claude Code Plugins That Actually Matter in 2025
The best Claude Code plugins right now are a mix of CLI tools, skills, and frameworks — not just MCPs. The ecosystem is moving fast, and most people are installing the wrong stuff. I test these tools constantly across client projects and my own builds, and these 10 are the ones that actually moved the needle. Here's exactly what they are, how to install them, and when to use each one.
The Claude Code add-on ecosystem is genuinely overwhelming right now. New MCPs, new CLIs, new skills, new frameworks — something drops every single day. And honestly, most of it is noise. So I wanted to cut through all of that and give you the 10 tools I reach for consistently. For each one, I'll cover why you should care, how to install it, and a real use case.
If you add even one of these to your workflow, you'll feel the difference immediately.
Why Should You Use CLIs Instead of MCPs in Claude Code?
Before we get into the list, I need to address something that tripped me up personally — and I do this full-time.
CLI tools are almost always better than MCP servers for Claude Code. Here's why: MCPs (Model Context Protocol servers) were created by Anthropic in 2024 as a standard for connecting AI assistants to external systems. That was useful when our AI tools lived in chat windows. But now? Claude Code lives in the terminal with us. It can use CLI tools directly, and CLIs are purpose-built for that environment.
MCPs are essentially trying to replicate what a CLI already does, but with extra overhead because they're not running natively in the terminal. So the rule is simple: if a tool has both a CLI and an MCP version, go with the CLI every time.
I was using the Supabase MCP for months before a YouTube comment called me out and pointed me toward the Supabase CLI. Me — someone who uses Supabase daily and understands that CLIs are superior. I just didn't know the CLI existed. So if you feel overwhelmed trying to keep up with this stuff, you're in good company.
With every CLI you install, also check if there's a matching skill available. Skills teach Claude Code how to use that specific CLI in the right way. Most good CLI tools ship with one.
What Is the Supabase CLI and How Do You Set It Up?
Tool number one: the Supabase CLI (not the MCP — I'll explain).
Supabase handles two things you'll eventually need in any serious project: databases and authentication. It's built on open-source tools like PostgreSQL, has a generous free tier, and you can self-host the entire stack if you want.
The installation changes depending on your OS, but you can find the exact commands in the Supabase docs. Or — and this is the beauty of Claude Code — just type: "Hey, download and install the Supabase CLI tool." Claude Code will handle the dependency installation for you.
Once installed, you'll want to add the Supabase skill. Search for it in the Claude Code plugin marketplace — it's listed as a Claude Code plugin. The skill tells Claude Code exactly how to interact with Supabase's CLI commands.
Use cases: Any project that needs a database (standard Postgres or vector databases for RAG), user authentication, profile management, or data separation between users. And you interact with all of it through plain language — no more hopping into the Supabase dashboard to write SQL manually.
What Is the Skill Creator Skill and Why Is It a Big Deal?
Tool number two is, in my opinion, the most powerful plugin on this entire list: the Skill Creator skill from Anthropic.
This isn't just a tool that creates skills. It lets you modify existing skills, measure their performance, and run A/B tests to see if your changes actually made things better. Anthropic published a full blog post about the improvements they made to this tool, and the implications are huge.
Before the Skill Creator, building skills was kind of a gut-feel exercise. Did the skill actually improve Claude Code's output? Maybe. Did you have data to prove it? Probably not. Now you can run blind A/B tests — with the skill vs. without it, current version vs. previous version — and get actual performance metrics.
To install it: Run /plugin inside Claude Code, search for "skill creator," and hit install. You'll see it listed as skill-creator.
To use it: Either run /skill-creator as a slash command, or just tell Claude Code in plain language: "Use the skill creator to build a new skill for [X]" or "Run an eval on my current skill." Under the hood, it's a text prompt — you can see the entire thing on Anthropic's GitHub skills repo.
This is the tool that turns skill-building from guesswork into engineering. If you're serious about improving Claude Code's output quality, start here.
How Does the GSD Framework Improve Claude Code Projects?
Tool number three: GSD (Get Shit Done), an orchestration framework that sits on top of Claude Code.
GSD changes how Claude Code operates when you're building projects from scratch. It enforces spec-driven development, breaking your project into phases and features with extreme rigor. But the real magic is in how it handles context.
One of the biggest problems with long Claude Code sessions is context rot — you're deep into a conversation, the context window is filling up, and output quality degrades. GSD solves this by spawning fresh sub-agent contexts for each task. Every new portion of the plan gets a clean context window, so you never hit that back-half degradation.
The GSD GitHub repo has the install command — it's a single line. Once installed, you start a new project with /gsd new-project, which kicks off what I'd describe as plan mode on steroids. It forces Claude Code to really nail down what you're building before writing a single line of code.
If you've built anything substantial with Claude Code, you know how critical the planning phase is. GSD makes that phase 10x more rigorous, and the results show in the final output.
How Do You Connect NotebookLM to Claude Code?
Tool number four: notebooklm-py, the CLI tool that connects Google's NotebookLM to Claude Code.
This is the backbone of my research workflows right now. Everything I can do inside NotebookLM — research, analysis, creating deliverables like podcasts, videos, infographics, slide decks, flashcards — I can now trigger from the Claude Code terminal. And there's no official NotebookLM API, so this tool is essentially reverse-engineering that access for us.
The value play here is wild. NotebookLM's capabilities are basically free, and now you can orchestrate them programmatically through Claude Code.
To install, head to the notebooklm-py GitHub, copy the install commands, and paste them into your terminal. After installation, run notebooklm in your terminal — it'll pop up a browser for you to log into your NotebookLM account.
Since this is a CLI tool, you'll want the skill too. The skill install command is included in the repo. Once everything's connected, you just talk to Claude Code in plain language: "Create a notebook with these sources and generate a podcast summary" — and it handles the rest.
Should You Use Obsidian with Claude Code?
Tool number five: Obsidian, but specifically in a personal assistant context.
Here's the thing — this isn't a one-size-fits-all recommendation. Obsidian plus Claude Code is great when you have vast, sprawling, accumulating context in the form of text documents and markdown files. Think personal knowledge management, not typical coding projects.
The setup is dead simple. No CLI to install, no skill needed. Just:
- Download Obsidian
- Create a folder designated as your vault
- Open Claude Code inside that folder
- Tell Claude Code: "When we create markdown files, follow Obsidian conventions"
That's it. Obsidian gives you visual insight into how your documents connect, keeps everything organized, and Claude Code handles the creation and management of those documents through natural language.
I use this for my own "second brain" setup, where I have a folder literally called "the vault." It's perfect for anyone using Claude Code as a personal assistant rather than purely as a coding tool.
How Do You Deploy Projects from Claude Code with Vercel?
Tool number six: the Vercel CLI.
Vercel is my go-to for deployments because it's simple and has a solid free tier. With the CLI installed, you can go from code to deployed project without ever leaving Claude Code.
Install with a single command (or just ask Claude Code to install it for you). Then grab the Vercel skill from Vercel's website — they actually have a bunch of skills available, some of which are pretty interesting. The deploy skill install is a single terminal command.
This tool pairs perfectly with Claude Code's agent loops. You can set up a loop that calls the Vercel CLI to check deployment statuses as you push changes, automating the entire code-to-production pipeline. Combined with the GitHub CLI (tool #8 on this list), you get a workflow where coding, pushing to GitHub, and deploying all happen from one place using natural language.
What Is the Playwright CLI and How Does It Work with Claude Code?
Tool number seven: the Playwright CLI for browser automation.
Playwright is an open-source tool from Microsoft, and it's become my preferred way to give Claude Code the ability to interact with the web. We're talking everything from automated testing to filling out forms to — yes — going to Amazon and filling up a shopping cart.
Head to the Playwright CLI GitHub, copy the install command, and paste it into your terminal. Skills install with playwright-cli install-skills. Done.
Use cases are broad:
- Testing UI design changes automatically
- Form submission testing
- Browser-based automations that would be painful to do manually
- Any repetitive browser task you're doing while developing
One thing worth calling out: the playwright-cli show command gives you a visual dashboard showing what all your browser agents are doing, even if they're running headless. Really useful when you have multiple automations running simultaneously.
Why Should Every Claude Code User Install the GitHub CLI?
Tool number eight: the GitHub CLI.
This one's straightforward. With the GitHub CLI installed, you can do everything you normally do on GitHub — repos, PRs, issues, pushes — directly from Claude Code using natural language.
Installation instructions are on the GitHub CLI page, broken down by operating system. Or just ask Claude Code to install it.
Here's the good news: coding agents like Claude Code are so deeply familiar with Git and GitHub that you don't really need a separate skill for this one. Just having the tool installed and talking to Claude Code in plain language is enough. But if you want one anyway, use the Skill Creator (tool #2) to build a custom one.
The real power move is combining the GitHub CLI with the Vercel CLI. Code → push to GitHub → deploy to Vercel, all from one terminal session, all through natural language. That's the workflow.
How Do You Use Firecrawl for Web Scraping in Claude Code?
Tool number nine: the Firecrawl CLI for web scraping.
Firecrawl's pitch is simple: web scraping where the data comes back tailor-made for AI agents. It's a significant step up from just asking Claude Code to use its built-in web search tool.
The install is one line — search "Firecrawl CLI tool" and you'll find it. The command also includes the skill install. Firecrawl has four core commands:
- Scrape — pull clean data from a single page
- Crawl — work through an entire site
- Map — get a site's structure
- Search — find and scrape relevant pages
If you're new to web scraping, those might sound like they overlap. But Claude Code knows exactly which command to use for which job — that's the whole point of the skill.
Standard use cases: competitor research, documentation ingestion, site change monitoring, and deep research with source citations. For anyone doing research-heavy work in Claude Code, Firecrawl is non-negotiable.
What Is the Excalidraw Diagram Skill for Claude Code?
Tool number ten: the Excalidraw Diagram Skill, created by Cole Medin.
This one is a recent addition to my stack and I'm already hooked. It lets you create diagrams through natural language with Claude Code. If you've ever used Excalidraw manually, you know how tedious it can be to build diagrams by hand. This skill eliminates all of that.
I used it to create all the diagrams for my recent video on the six levels of Claude Code. Every single chart, every visual breakdown — none of it was done manually. I pointed Claude Code at my notes, said "break down each level into a visual chart," went back and forth in plan mode to refine the concepts, and the skill generated everything.
To install, head to the Excalidraw Diagram Skill GitHub, clone the repo, and copy it into your project's .claude/skills/ directory. Then use it through natural language like everything else.
Best use case: Point Claude Code at a body of knowledge or documentation, then ask it to create visual representations. Presentations, explainer diagrams, architecture charts — anything where you'd normally spend 30+ minutes dragging boxes around in a diagramming tool.
Shout out to Cole Medin for building this. If you're not following his YouTube channel, you should be.
The Right Number of Plugins Is Not "All of Them"
Look, there's a trap I see people fall into: thinking that more tools equals better output. It doesn't. Increased capability does not automatically equal increased performance. There's a sweet spot, and loading up Claude Code with every plugin you can find will actually hurt you through context bloat and decision paralysis.
Pick the tools that match your actual workflow. If you're not doing web scraping, skip Firecrawl. If you're not making diagrams, skip Excalidraw. The goal is finding the middle ground where each tool is pulling its weight.
My recommendation for most people getting started: install the Supabase CLI, the Skill Creator, and either the Vercel or GitHub CLI depending on your deployment needs. That covers 80% of what you'll need. Add the rest as specific projects demand them.
Frequently Asked Questions
What's the difference between Claude Code plugins, skills, and MCPs?
Plugins are bundled packages that can include skills, hooks, and other extensions. Skills are instruction files that teach Claude Code how to perform specific tasks. MCPs (Model Context Protocol servers) are external connections to third-party services. For most use cases in 2025, CLI tools paired with skills outperform MCPs because they run natively in the terminal where Claude Code lives.
Do I need coding experience to use these Claude Code plugins?
No. Every tool on this list is operated through natural language — you tell Claude Code what you want in plain English, and it handles the technical execution. Installation usually requires copying a single command into your terminal. Claude Code can even install most of these tools for you if you just ask.
How many Claude Code plugins should I install at once?
Start with 2-3 that match your immediate workflow needs. Adding too many tools creates context bloat and can actually degrade Claude Code's performance. The Skill Creator and Supabase CLI are solid starting points for most developers. Add specialized tools like Firecrawl or Playwright only when a specific project requires them.
Are these Claude Code plugins free to use?
Most of them are free or have generous free tiers. Supabase, Vercel, and GitHub all offer free tiers that cover significant usage. Firecrawl has a free tier as well. The Skill Creator, GSD framework, Playwright CLI, Obsidian, and Excalidraw Diagram Skill are all free. NotebookLM is free through Google.
Should I use the Supabase MCP or the Supabase CLI with Claude Code?
Always use the Supabase CLI over the MCP. The CLI runs natively in the terminal and has less overhead than the MCP, which adds an extra abstraction layer. This applies broadly — whenever a tool offers both a CLI and an MCP version, choose the CLI. Pair it with the appropriate skill from the plugin marketplace so Claude Code knows exactly how to use it.
If you want to go deeper into Claude Code plugins and AI development workflows, join the free Chase AI community for templates, prompts, and live breakdowns. And if you're serious about building with AI, check out the paid community, Chase AI+, for hands-on guidance on how to make money with AI.
